Add mistral-large-4

Add Aleph-Alpha/Kolibri-1

Add gemini-4-argon

Add XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B; scope Pro/Flash-only benchmarks

Add XiaomiMiMo/MiMo-V2.6-Flash-RL

Add XiaomiMiMo/MiMo-V2.6-Pro-RL

Add qwen3.8-omni-flash

Add deepseek-ai/DeepSeek-V4.1-Flash and nex-agi/Nex-N2.5-Pro

Add muse-spark-1.3

Add gpt-6-astra

Add claude-fable-5-1 and claude-mythos-5-1

Add ibm-granite/granite-4.2-30b and deepseek-ai/DeepSeek-V4-Flash-Vision-Exp

Add tencent/Hy4-preview

Add Qwen/Qwen3.8-Flash-Next

Update benchmarks row count to 158

Add metadata for 9 Rare+ benchmarks (Toolathlon-Verified, AutomationBench, ExploitBench, FrontierSWE, Finance Agent v2, ExploitGym, JobBench, SWE Marathon, GDPval-AA v2)

Update benchmarks row count to 149

Add DeepSWE hf_dataset, catalog DeepSWE + Agent's Last Exam

Remove low-confidence hf_dataset matches for TAU2-Bench, Terminal-Bench 2.1, RULER, BrowseComp

Fix regression: flatten models.benchmarks back to plain name strings

Add hf_dataset column to benchmarks table

Add Ornith-1.5 family (9B, 35B-A3B, 397B)

Canonicalize GLM-5.3 ProgramBench Almost Solved -> ProgramBench

Add zai-org/GLM-5.3

Add dots-studio/dots3-note-prev

Add 9 benchmarks from dots3-note-prev: ARC-AGI-3, VibeLifeBench, LiveBrowseComp, VibeSearchBench, WorldVQA, CharXiv-RQ, MME Video-V2, VideoZeroBench, MMAU-Pro

Add 8 benchmarks from gemini-3.7-flash: FrontierCode 1.1, Code Arena, Terminal-Bench 3.0, GDP.pdf, CharXiv w/ Tools, BioMysteryBench (2 splits), LAB-Bench 2

Add gemini-3.7-flash

Add DeepSeek-V4-Pro-0813

Fix Qwen3.8-2.4T-A95B: NL2Repo-Bench -> canonical NL2Repo

Fix duplicate: NL2Repo-Bench merged into existing NL2Repo entry

Add 6 benchmarks from Qwen3.8-2.4T-A95B: DeepSWE 1.1, NL2Repo-Bench, PLawBench, PRBench-Legal, PRBench-Finance, $OneMillion-Bench

Add Qwen/Qwen3.8-2.4T-A95B

Clarify benchmarks split has partial coverage of referenced names

Restore all benchmark names in models.benchmarks (flatten to strings, no filtering)

Document flattened benchmarks field and benchmarks split

Flatten models.benchmarks to name strings matching benchmarks split

Add benchmarks config (benchmarks.jsonl split) to dataset card

Add benchmarks split: 125 benchmark metadata entries

Add nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16

Update README for simplified models schema

Simplify models schema: inline source link and benchmarks, drop source_id/leaderboards

Add Motif-Technologies/Motif-3

Add Motif-Technologies/Motif-3

Add meta-models/Muse-Glimmer-30B

Add meta-models/Muse-Glimmer-30B

Add A.X-K2, DeepSeek-V4-Flash-0731, MiniMax-H3 sources

Add A.X-K2, DeepSeek-V4-Flash-0731, MiniMax-H3

Add Kimi-K3, Inkling, LFM2.5-1.2B-Instruct sources

Add Kimi-K3, Inkling, LFM2.5-1.2B-Instruct