[ { "Model": "Kimi K2.7 Code abliterated", "Uploader / repo": "huihui-ai Huihui-Kimi-K2.7-Code-abliterated (GGUF copy pulled from loxxxxx/Huihui-Kimi-K2.7-Code-abliterated-GGUF)", "Quant": "UD-Q2_K_XL (experts IQ2_XS/IQ3_XXS/Q4_0; 42 shared tensors MXFP4)", "File size": "316 GB, 8 shards (v100 note says 339 GB)", "Hardware / loader": "Mac Studio M3 Ultra 512 GB, llama.cpp Metal; -c 131072..262144 --parallel 2", "Gen tok/s": "~24 single stream (stock build 22.2 @2k; deployed fast build 23.8 @2k, 11.2 @75k; exact candidate13 30.6 @2k, 17.57 @55k, not deployed); real SWE agent traffic 3.4 avg (old build) / 10.1 avg (fast build), 2 slots", "Prompt tok/s": "184 / 146 / 90 @2k/8k/26k (fast build); round 7 exact: 3.7k 181->254, 26k 102->120", "Refusals /8": "0/8", "SWE-bench Verified 100": "A: 81 (71 first full pass Nov 2 at 64k ctx + 10 from ONE retry of each of the 29 failures at larger ctx, Nov 9-11; union of 12 report files). Single pass = 71.", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "16x V100 (2 V100 servers) llama.cpp RPC over 1 Gb: 13.2 tok/s 1 lane, 39.0 total at 100 lanes, refusal probe 0 (7 valid, 1 timeout); patched v6b build 20.0 tok/s 1 chat, 84.3 total at 100 chats, 2.6k prompt 236 tok/s. Router tool battery 14/14. House bug_100 harness (network on, not SWE-bench): 73/99. A 2026-11-01 Mac run scored 0/99 because the server died (void).", "Verdict": "LIVE (Mac Studio; deep tier of our house build Auto", "Date": "2026-11-02 .. 2026-11-11 (SWE); speeds to 2026-11-14" }, { "Model": "Kimi K2.7 Code (NOT abliterated)", "Uploader / repo": "unsloth/Kimi-K2.7-Code-GGUF", "Quant": "UD-IQ1_M (1-bit dynamic)", "File size": "283 GiB (size from HF listing in sizing note)", "Hardware / loader": "rented Vast 4x H200, llama.cpp -c 393216 --parallel 6", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "A: 68 (single pass; 4 empty incl. 3 ContextWindowExceeded at 128k)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "Pod destroyed after test (rental)", "Date": "2026-11-01" }, { "Model": "Kimi K2.7 Code custom quant 'kimi-abl-Q4C408' (named abliterated, behaves fully aligned)", "Uploader / repo": "in-house/in-house bake (source not recorded)", "Quant": "Q4C408 = Q4_K_M base, experts L0-30 Q3_K + L31-60 Q2_K (3.06 BPW, not 4-bit)", "File size": "392 GB (365 GiB)", "Hardware / loader": "Mac Studio M3 Ultra llama.cpp Metal; also A100 server CPU-only and CPU+2xA100 split", "Gen tok/s": "Mac median 5.9, fastest 12-13.7 (sizing note; another note says ~17); A100 server CPU 1.3 (warm); A100 server split 2.12 on a new prompt", "Prompt tok/s": "A100 server CPU ~2; A100 server split 2.43", "Refusals /8": "8/8 (harness printed a false 0% because refusals were >700 chars)", "SWE-bench Verified 100": "A: 7 (90 empty patches; a session log partial read 12/100)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "DELETED everywhere 2026-11-05/06 (useless: refuses, empty patches)", "Date": "2026-11-05 .. 2026-11-08" }, { "Model": "Kimi K2.7 Code abliterated, 'local Q2' (exact build not recorded; predates the huihui UD-Q2_K_XL download of 2026-11-05)", "Uploader / repo": "", "Quant": "Q2 (unspecified)", "File size": "", "Hardware / loader": "local (Mac, per ledger)", "Gen tok/s": "24.1 (session log)", "Prompt tok/s": "", "Refusals /8": "0/8 (2 empty probes)", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "149", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "dropped after HumanEval", "Date": "2026-10-01" }, { "Model": "Kimi K3 (official, hosted)", "Uploader / repo": "Moonshot via Kimi Code CLI subscription (model id k3)", "Quant": "", "File size": "", "Hardware / loader": "hosted (Kimi Code CLI as agent)", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "B: 91 (stitched: 65 of first 72 on acct A with web tools ON + 24 of last 28 + 2 of 3 pytest re-runs; >=3 bugs used future git history, clean re-runs k3c still solved them). C (Oct 25 'Kimi reference', 81): Oct 28 note records it as K3, our scoreboard says model unconfirmed", "216-task coder harness": "", "HumanEval /164": "162", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "Too big to run locally at usable quant (~1.5 TB at Q4)", "Verdict": "reference only", "Date": "2026-10-01 (HumanEval); 2026-10-25 (ref 81); 2026-11-09..10 (B 91)" }, { "Model": "Kimi K3", "Uploader / repo": "unsloth/Kimi-K3-GGUF", "Quant": "UD-Q1_0 = IQ1_XXXS (ggml type 66, 1.19 bpw)", "File size": "466 GB", "Hardware / loader": "Mac Studio M3 Ultra 512 GB, unslothai llama.cpp PR #61 iq1-narrow, --device none (CPU only; no Metal kernels)", "Gen tok/s": "2.3", "Prompt tok/s": "4.8", "Refusals /8": "0/8 (valid 8)", "SWE-bench Verified 100": "", "216-task coder harness": "not run (600 s task cap infeasible)", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "kept on disk; replaced as served 'K3' by REAP576", "Date": "2026-10-19" }, { "Model": "Kimi K3 REAP576 (576/896 experts pruned)", "Uploader / repo": "hellohazime/Kimi-K3-REAP-512GB-GGUF", "Quant": "REAP576-IQ2_XXS (2-bit)", "File size": "478.5 GB", "Hardware / loader": "Mac Studio M3 Ultra 512 GB, llama.cpp PR #48 build, full Metal -ngl 99 --cache-reuse 0 -c 8192", "Gen tok/s": "2.8-3.1", "Prompt tok/s": "61 (2,449-tok prompt); 11 (37-tok prompt)", "Refusals /8": "not probed", "SWE-bench Verified 100": "", "216-task coder harness": "not run (we: pointless at ~3 tok/s)", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "we 2026-10-19: stop tests, keep the model", "Date": "2026-10-18/19" }, { "Model": "GLM-5.3-Flash abliterated", "Uploader / repo": "huihui-ai/Huihui-GLM-5.3-Flash-abliterated-GGUF", "Quant": "UD-Q6_K_XL", "File size": "291.8 GB (273 GiB), 7 shards", "Hardware / loader": "Mac Studio M3 Ultra 512 GB (Mac Studio; Mac Studio for ctx re-run), llama.cpp Metal (fast build ff5a233, then glm-opt candidate1..7)", "Gen tok/s": "22.66 (2026-11-07; later called mislabeled/slow build); ~24 fast build; candidate1 25.5 plain, 21.4 @64k, 20.7 @100k, 43 total 4 slots @64k; Mac Studio baseline 25.2 @2k / 22.0 @100k; 29.23-29.26 in round with candidate7", "Prompt tok/s": "29.6 (2026-11-07); candidate3 397 (exact -ub 2048 split; earlier quoted 341->410 was withdrawn)", "Refusals /8": "0/8 (reasoning model: needs >=1500 tokens or reads as empty/refused)", "SWE-bench Verified 100": "A: 80 (79 first pass at 131k ctx x4 slots, drafting on, text mode + 1 from 262k ctx re-run of its 2 ctx exits). 0 step limits, 2 format errors", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "LIVE (Mac Studio); our house build 4.1", "Date": "2026-11-07 (probe/speed); 2026-11-11/12 (SWE); speed rounds to 2026-11-14" }, { "Model": "GLM-5.3-Flash (not abliterated)", "Uploader / repo": "unsloth GGUF", "Quant": "UD-Q4_K_XL (session log)", "File size": "186 GB (ledger); ~200 GB (session log)", "Hardware / loader": "Mac Studio, llama.cpp with PR #27754 (glm5next arch)", "Gen tok/s": "21.5", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "Reasoned endlessly: 11,908 chars of thinking and no answer on a trivial prompt with 4,000 tokens, even with thinking disabled", "Verdict": "DELETED on we order 2026-10-02 23:19 (could not be scored)", "Date": "2026-10-02" }, { "Model": "MiMo-V2.6-Flash-MOPD heretic", "Uploader / repo": "mradermacher/MiMo-V2.6-Flash-MOPD-heretic-i1-GGUF", "Quant": "i1-Q4_K_M", "File size": "186.7 GB (186,657,491,872 B; 4 raw parts cat'd)", "Hardware / loader": "Mac Studio M3 Ultra 512 GB, llama.cpp mainline Metal, --reasoning-budget 0 (thinking off); later mimo-opt candidate2 (attention lever + lookup drafting)", "Gen tok/s": "49.1 (156-tok, 2026-10-19); speed work: 51 short, 44.0 @64k, 40.9 @100k, 15.5/stream with 4 streams @64k", "Prompt tok/s": "71 (short prompt)", "Refusals /8": "0/8 (valid 8)", "SWE-bench Verified 100": "A: 66 (65 first pass at 131k x4, drafting on, thinking off + 1 from 262k re-run of 15 ctx exits; MiMo loops on large file reads)", "216-task coder harness": "152/213 = 71.4% (host mode, jobs=2, 900 s/task; 49 step-limit stalls, 2 timeouts, 11 tamper)", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "kept on disk; not added to router (we 2026-11-14); was the Mac's served model after 2026-10-19", "Date": "2026-10-19 (213); 2026-11-11/12 (SWE)" }, { "Model": "MiMo-V2.6-Flash-MOPD Yamz-Uncensored", "Uploader / repo": "spiritfather/MiMo-V2.6-Flash-MOPD-Yamz-Uncensored-GGUF", "Quant": "Q4_K_M", "File size": "189.5 GB (189,508,168,096 B)", "Hardware / loader": "Mac Studio M3 Ultra, llama.cpp mainline Metal -ngl 99 -c 65536 -np 2, --reasoning-budget 0", "Gen tok/s": "43.9 (200-tok)", "Prompt tok/s": "509 (2,436-tok prompt); 62 (37-tok)", "Refusals /8": "1/8 (identical on 3 runs at temp 0) -> zero-refusal bar FAIL", "SWE-bench Verified 100": "", "216-task coder harness": "not run (gated on 0 refusals)", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "replaced by heretic build", "Date": "2026-10-18" }, { "Model": "MiMo-V2.6-Flash-MOPD heretic", "Uploader / repo": "mradermacher (same repo)", "Quant": "i1-Q2_K", "File size": "105.5 GiB", "Hardware / loader": "Mac Studio M3 Ultra, llama.cpp -ngl 99 -c 65536 -np 2 --reasoning-budget 0", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "our house build Auto router test as front-desk screener: SAFETY 9/12 (made answers up on 3); tool battery 14/14; 14 command tasks in 467 s (our house build 3.0 247 s, our house build 2.0 424 s); first token 0.3-0.5 s", "Verdict": "failed router screener test; our house build 2.0 put back", "Date": "2026-11-06" }, { "Model": "DeepSeek-V4.1-Flash-1017 abliterated (763B total / 16B active)", "Uploader / repo": "mrfakename/DeepSeek-V4.1-Flash-1017-abliterated-GGUF", "Quant": "Q4_K_M", "File size": "444.7 GB, 10 shards", "Hardware / loader": "Mac Studio M3 Ultra, llama.cpp PR #28696 branch (ds41 @2076412) Metal, thinking off, 12,288 tok/slot (16K/slot crashes)", "Gen tok/s": "27.3 (160-tok, 8K/slot)", "Prompt tok/s": "317 (2,432-tok)", "Refusals /8": "0/8 (valid 8)", "SWE-bench Verified 100": "", "216-task coder harness": "136/213 = 63.8% (18 timeouts, 11 context-cap failures, 27 stalls; paired vs MiMo exact McNemar p=0.007)", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "DELETED 2026-10-19 (we: remove; MiMo scored higher)", "Date": "2026-10-19" }, { "Model": "DeepSeek-V4.1-Flash-1017 (official, not abliterated)", "Uploader / repo": "ddalcu/DeepSeek-V4.1-Flash-1017-MLX-Serve-Lossless", "Quant": "MLX native (routed experts FP4/MXFP4, trunk FP8)", "File size": "477 GB", "Hardware / loader": "Mac Studio M3 Ultra (Mac Studio) solo, mlx-serve --ctx-size 8192", "Gen tok/s": "13.0", "Prompt tok/s": "78", "Refusals /8": "8/8", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "DELETED (official baseline only)", "Date": "2026-11-07" }, { "Model": "DeepSeek-V4.1-Flash-1017 OBLITERATUS abliteration, re-serialized by Kimi to llama.cpp layout (obl-v41)", "Uploader / repo": "OBLITERATUS/...-1017-obliterated (BF16 1.4 TB source)", "Quant": "MXFP4/Q8 GGUF (obl-v41.gguf)", "File size": "648 GB layer weights (older obl-v41-mxfp4.gguf ~699 GB)", "Hardware / loader": "both Mac Studios: Mac Studio client + Mac Studio ggml-rpc-server over Thunderbolt", "Gen tok/s": "2.5 (Kimi's CPU config); Metal (MTL0) layout result pending", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "not scored ( config exists on RTX PRO 6000 box, no report file)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "pending (we queue: last)", "Date": "2026-11-07/08" }, { "Model": "DeepSeek-V4-Flash-0731 abliterated", "Uploader / repo": "", "Quant": "4-bit", "File size": "165 GB", "Hardware / loader": "Mac Studio", "Gen tok/s": "27 (session log)", "Prompt tok/s": "", "Refusals /8": "0/8", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "26-task real-task suite: 2-3/26 (session log; suite non-discriminating)", "Verdict": "not adopted", "Date": "2026-09-29/30" }, { "Model": "Qwen3.8-27B abliterated", "Uploader / repo": "huihui-ai/Huihui-Qwen3.8-27B-abliterated", "Quant": "bf16 (213 harness, vLLM A100); FP8-on-the-fly vLLM 0.32 (RTX PRO 6000 box SWE runs from 2026-10-24); production GGUF: UD-Q4_K_XL 17 GB (2026-10-18 note) / Q8_0 29.0 GB on RTX PRO 6000 box (file dated 2026-10-26)", "File size": "Q8_0 29,047,084,320 B; UD-Q4_K_XL 17 GB", "Hardware / loader": "A100 80GB vLLM (213); RTX PRO 6000 vLLM (SWE); RTX PRO 6000 llama.cpp (prod)", "Gen tok/s": "28 (exam 2, Mac); Mac 27 llama.cpp / 31 MLX (09-29); ~60 UD-Q4_K_XL RTX (10-18); 46.9 Q8_0 RTX PRO 6000 box (10-27); 37-41 RTX PRO 6000 box (11-09); 14.9 RTX PRO 6000 box card (11-01); 38 idle / ~20 / 14 under shared load RTX PRO 6000 box (11-14)", "Prompt tok/s": "", "Refusals /8": "0/8 (valid)", "SWE-bench Verified 100": "A: 63 per our scoreboard ('64 raw'; report file has 64), net-off, Nov 1. Net-ON mini-swe-agent (contaminated): 79 (Oct 24-25), 69 re-run (Oct 29). 66-bug subset: net-off 41/66, net-on 53/66. house-agent CLI runs: 49, 53, 52, 38; 66-set prompt variants 38/35/34", "216-task coder harness": "150/213 = 70.4% (2026-09-14, bf16, host mode). Re-measured 2026-10-24 with vLLM: base 157/213; eval_repos 29: 20-23/29", "HumanEval /164": "155", "HumanEval+ / MBPP+ / LiveCodeBench": "HumanEval+ 147/164, MBPP+ 295/378, LiveCodeBench 275/454 (60.6%), LCB hard 39.4%", "Other results": "Router tool battery 14/14; 14 command tasks in 247 s", "Verdict": "LIVE (RTX PRO 6000 box, standard tier); held off the top SWE board by us 2026-11-10", "Date": "2026-09-14 .. 2026-11-14" }, { "Model": "Qwen3.8-27B abliterated + our house build LoRA r1 (1 epoch QLoRA on 466 self-generated passing tasks)", "Uploader / repo": "in-house LoRA on huihui-ai/Huihui-Qwen3.8-27B-abliterated", "Quant": "bf16 base + LoRA adapter, served by vLLM (FP8 on the fly) on RTX PRO 6000 box", "File size": "", "Hardware / loader": "RTX PRO 6000 96 GB, vLLM 0.32", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "net-ON mini-swe-agent: 22/38 (partial run, report covers 38 submitted)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "213 harness: 154/213 (base 157); eval_repos 28/29 (base 23/29). Integrity note: 'r1 cratered the model (honest ~20)'", "Verdict": "not shipped (SFT recipe stopped 2026-10-31: no honest gain over base)", "Date": "2026-10-24" }, { "Model": "Qwen3.8-27B abliterated + our house build LoRA r2 (466 self + 413 Kimi-session + 150 other examples)", "Uploader / repo": "in-house LoRA on huihui-ai/Huihui-Qwen3.8-27B-abliterated", "Quant": "bf16 base + LoRA adapter, served by vLLM (FP8 on the fly) on RTX PRO 6000 box", "File size": "", "Hardware / loader": "RTX PRO 6000 96 GB, vLLM 0.32", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "net-ON mini-swe-agent: 66 (18 LimitsExceeded)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "not shipped (SFT recipe stopped 2026-10-31: no honest gain over base)", "Date": "2026-10-25" }, { "Model": "Qwen3.8-27B abliterated + our house build LoRA r3 (rejection-sampled SWE-rebench trajectories)", "Uploader / repo": "in-house LoRA on huihui-ai/Huihui-Qwen3.8-27B-abliterated", "Quant": "bf16 base + LoRA adapter, served by vLLM (FP8 on the fly) on RTX PRO 6000 box", "File size": "", "Hardware / loader": "RTX PRO 6000 96 GB, vLLM 0.32", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "net-ON mini-swe-agent: 78 (base 79)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "not shipped (SFT recipe stopped 2026-10-31: no honest gain over base)", "Date": "2026-10-26" }, { "Model": "Qwen3.8-27B abliterated + our house build LoRA r4 (r3 + retries of 441 misses, 912 ex)", "Uploader / repo": "in-house LoRA on huihui-ai/Huihui-Qwen3.8-27B-abliterated", "Quant": "bf16 base + LoRA adapter, served by vLLM (FP8 on the fly) on RTX PRO 6000 box", "File size": "", "Hardware / loader": "RTX PRO 6000 96 GB, vLLM 0.32", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "net-ON mini-swe-agent: 78; net-OFF on 66-bug subset: 37/66 (base 41/66)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "not shipped (SFT recipe stopped 2026-10-31: no honest gain over base)", "Date": "2026-10-27 / 2026-10-30" }, { "Model": "Qwen3.8-27B abliterated + our house build LoRA r5 (672 cheat-scrubbed examples)", "Uploader / repo": "in-house LoRA on huihui-ai/Huihui-Qwen3.8-27B-abliterated", "Quant": "bf16 base + LoRA adapter, served by vLLM (FP8 on the fly) on RTX PRO 6000 box", "File size": "", "Hardware / loader": "RTX PRO 6000 96 GB, vLLM 0.32", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "net-OFF on 66-bug subset: 36/66 (base 41/66)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "not shipped (SFT recipe stopped 2026-10-31: no honest gain over base)", "Date": "2026-10-31" }, { "Model": "Qwen3.5-27B abliterated", "Uploader / repo": "huihui-ai (Huihui-Qwen3.5-27B-abliterated)", "Quant": "bf16", "File size": "", "Hardware / loader": "A100 SXM4 80 GB, vLLM 0.28.0, host-mode harness", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "1/8 -> zero-refusal bar fail", "SWE-bench Verified 100": "", "216-task coder harness": "148/213 = 69.5% (vs 3.8-27B: 9 vs 7 wins, p=0.80)", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "rejected (refusal)", "Date": "2026-09-14" }, { "Model": "Qwen3-Coder-30B-A3B-Instruct abliterated", "Uploader / repo": "huihui-ai (Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated)", "Quant": "bf16 (213 harness); Q4_K_M GGUF in production", "File size": "Q4_K_M 18 GB", "Hardware / loader": "A100 80 GB vLLM bf16 (213); RTX PRO 6000 llama.cpp (prod)", "Gen tok/s": "112 Mac llama.cpp, 88 Mac MLX; 109 in 09-30 shootout (session log)", "Prompt tok/s": "", "Refusals /8": "0/8 (valid)", "SWE-bench Verified 100": "", "216-task coder harness": "87/213 = 40.8% (101 stalls, 19 timeouts; RTX PRO 6000 box says control ABORT, no valid score); earlier base run 62/216 = 28.7%", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "26-task suite 2-3/26 (09-30). Router screener re-test 2026-11-14 (Q4_K_M): SAFETY 19/19, HELP 14/18, first words 0.2 s", "Verdict": "LIVE as fast-lane screener (RTX PRO 6000 box", "Date": "2026-09-14; 2026-11-13/14" }, { "Model": "Qwen3-Coder-Next abliterated", "Uploader / repo": "huihui-ai; GGUF by bartowski/huihui-ai_Qwen3-Coder-Next-abliterated-GGUF", "Quant": "Q8_0", "File size": "85 GB, 3 shards (2026-11-06 note says 79 GB for the Mac copy)", "Hardware / loader": "Mac Studio llama.cpp (original; -c 131072 --parallel 2); RTX PRO 6000 llama.cpp from 2026-10-18 (-c 32768 -np 2 -fa on, q8_0 KV)", "Gen tok/s": "66 (exam 2; session log 65.6 on the Mac llama.cpp); ~7 on the Mac while our house build 4.0 benchmarks run (2026-11-06)", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "151", "HumanEval+ / MBPP+ / LiveCodeBench": "HumanEval+ 147/164, MBPP+ 291/378, LiveCodeBench 250/454 (55.1%), LCB hard 25.3%", "Other results": "Router tool battery 14/14; 14 command tasks in 424 s", "Verdict": "our house build 2.0 (speed role) 2026-10-02 .. 2026-11-13; replaced on RTX PRO 6000 box by 30B-A3B Q4 for VRAM", "Date": "2026-10-01/02; 2026-11-08" }, { "Model": "Qwen3-Coder-Next abliterated", "Uploader / repo": "huihui-ai weights; GGUF Huihui-Qwen3-Coder-Next-abliterated.Q6_K.gguf (session log says via mradermacher)", "Quant": "Q6_K", "File size": "62G per transfer log (session log: 65.5 GB)", "Hardware / loader": "RTX PRO 6000 96 GB, portable llama.cpp (qwen3next), n_ctx_slot 256000 x2", "Gen tok/s": "134-159 (serve log, short replies); session log quotes 29.1 and later says clean tok/s was never captured", "Prompt tok/s": "1,246-1,645 (2-3k prompts); 467-985 (short)", "Refusals /8": "0/8 (session log)", "SWE-bench Verified 100": "A: 47 (Nov 8, network off; 11 empty, 17 errors). Scoreboard calls it 'qwen Q6 run'; identity from A100 server + RTX PRO 6000 box serve log", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "loser, deleted (session log)", "Date": "2026-11-08" }, { "Model": "Qwen3-Coder-Next abliterated", "Uploader / repo": "huihui-ai weights; GGUF Huihui-Qwen3-Coder-Next-abliterated.Q8_0.gguf single file (session log says via mradermacher)", "Quant": "Q8_0", "File size": "79G per transfer log (session log: 84.8 GB)", "Hardware / loader": "RTX PRO 6000 96 GB, llama.cpp, n_ctx_slot 256000 x2 (512k for the patch run)", "Gen tok/s": "~139.5 (serve log, 74-tok reply); session log quotes 34.6 and later says clean tok/s was never captured", "Prompt tok/s": "1,443-1,681 (1.4-1.6k prompts)", "Refusals /8": "0/8 (valid 8, errors 0; first gate attempt unreachable 0 valid/8 errors)", "SWE-bench Verified 100": "A: 49 (Nov 8, network off; 61 good preds from a 76-bug partial + patch run of 17 ctx failures and missing bugs at 512k). Scoreboard calls it 'qwen Q8 run'", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "loser, deleted (session log); 2026-11-10 board mislabels Q8/Q6 as Qwen3-Coder 480B", "Date": "2026-11-08" }, { "Model": "Qwen3-Coder-480B-A35B-Instruct abliterated", "Uploader / repo": "huihui-ai (Huihui-Qwen3-Coder-480B-abliterated)", "Quant": "Q4_K_M GGUF", "File size": "270 GB, 32 shards", "Hardware / loader": "Mac Studio (Mac Studio) llama.cpp, 4 slots x 64k (SWE); local Q4 (exams)", "Gen tok/s": "25.9 (MODEL_TRACKER); 25 (exam 2); 25.5 (session log)", "Prompt tok/s": "", "Refusals /8": "0/8", "SWE-bench Verified 100": "A: 46 (87 completed, 41 unresolved, 12 patch-apply errors, 1 empty)", "216-task coder harness": "", "HumanEval /164": "156", "HumanEval+ / MBPP+ / LiveCodeBench": "HumanEval+ 146/164, MBPP+ 295/378, LiveCodeBench 211/454 (46.5%), LCB hard 15.8%", "Other results": "", "Verdict": "DELETED from both Macs 2026-11-07 (weak coder)", "Date": "2026-10-01/02; 2026-11-07" }, { "Model": "Qwen3.8-Flash-Next abliterated", "Uploader / repo": "huihui-ai (Huihui-Qwen3.8-Flash-Next-abliterated)", "Quant": "", "File size": "", "Hardware / loader": "rented 4x H100 (~$12)", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "138/213 (27B wins head-to-head 19-7; 59 step-limit, 27 format errors)", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "lost to our house build 3.0", "Date": "2026-10-05" }, { "Model": "Qwen3.8-Flash-Next (censored build; the abliterated download never finished)", "Uploader / repo": "", "Quant": "4-bit", "File size": "111 GB", "Hardware / loader": "Mac Studio", "Gen tok/s": "39 (session log)", "Prompt tok/s": "", "Refusals /8": "1/8 (session log) -> zero-refusal bar fail", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "26-task real-task suite: 4-5/26 (session log; band 2-5/26, non-discriminating)", "Verdict": "rejected", "Date": "2026-09-29/30" }, { "Model": "Qwen3.5-397B-A17B abliterated", "Uploader / repo": "huihui-ai/Huihui-Qwen3.5-397B-A17B-abliterated-GGUF", "Quant": "Q3_K", "File size": "189 GB, 21 shards", "Hardware / loader": "Mac Studio", "Gen tok/s": "32 (session log)", "Prompt tok/s": "", "Refusals /8": "1/8 (session log) -> zero-refusal bar fail", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "26-task real-task suite: 3/26 (session log); never run on the 213 harness", "Verdict": "rejected", "Date": "2026-09-29/30" }, { "Model": "Qwen3.8-27B MPOA heretic", "Uploader / repo": "", "Quant": "Q6_K", "File size": "22.08 GB (22,082,530,816 B)", "Hardware / loader": "A100 server, one A100 40 GB (llama.cpp CUDA)", "Gen tok/s": "32.9-34.8", "Prompt tok/s": "1,003.2 (22,767-token prompt); 797.9 (687-tok)", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "speed check only", "Date": "2026-11-05" }, { "Model": "Ornith-1.5-122B uncensored", "Uploader / repo": "(uploader not recorded; served as ornith:122b-uncensored)", "Quant": "NVFP4 (B200 run); Q4_K_M GGUF (Mac/RTX PRO 6000 box bake-off attempt, no report)", "File size": "Q4_K_M 79 GB (session log); 147 GB archive folder", "Hardware / loader": "rented Vast B200, vLLM, 131k ctx (run 1) / 256k ctx (re-run of 22)", "Gen tok/s": "~51 single user on the Mac; ~22 per task with 6 parallel (session log)", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "A-harness but NETWORK ON (stock, no; 10 of 100 runs downloaded fixes per session log): run 1 = 63 per notes / 64 in out_ornith_b200 report; combined = 67 per notes and archive README / 70 in out_ornith_combined report", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "set aside; archived to server (~$36 Vast)", "Date": "2026-10-27/28" }, { "Model": "MiniMax M3 (official)", "Uploader / repo": "", "Quant": "UD-Q4_K_XL (session log)", "File size": "~265 GB on the Mac / 247 GB on the rental (session log)", "Hardware / loader": "Mac Studio llama.cpp; rented 4x RTX PRO 6000 (box destroyed)", "Gen tok/s": "26 (ledger); 26.3 on the Mac, ~22 aggregate at 24 parallel (session log)", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "HumanEval+ 144/164, MBPP+ 290/378, LiveCodeBench not finished (~14 h projected)", "Other results": "", "Verdict": "not worth adopting; rental destroyed", "Date": "2026-10-02/03" }, { "Model": "Claude Fable 5.2 (hosted)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "Claude Code CLI as agent (claude -p, WebSearch/WebFetch disallowed)", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "B: 93 (0 future-history uses found)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "reference", "Date": "2026-11-10" }, { "Model": "Claude Opus 5.5 (hosted)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "Claude Code CLI as agent", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "B: 92. C: 94 (Oct 25 'Claude reference', recorded as Opus 5.5 in the Oct 28 note; removed from the net-off board by us 2026-11-10)", "216-task coder harness": "", "HumanEval /164": "164", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "reference", "Date": "2026-10-01; 2026-10-25; 2026-11-10" }, { "Model": "Claude Opus 5 (hosted)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "Claude Code CLI as agent (account B, RTX 5070 box)", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "B: 92 (2 bugs used future git history; clean re-runs still solved)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "reference", "Date": "2026-11-10" }, { "Model": "Claude Sonnet 5.5 (hosted)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "Claude Code CLI as agent (account B, RTX 5070 box)", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "B: 80", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "reference", "Date": "2026-11-10" }, { "Model": "Claude Opus 4.8 (hosted)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "Claude Code CLI as agent", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "B: 79 (2 bugs used future git history; clean re-runs still solved)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "reference", "Date": "2026-11-10" }, { "Model": "OpenAI Codex gpt-6-astra (hosted)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "Codex CLI, effort medium, -s workspace-write (OS-sandboxed, no network)", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "B: 83", "216-task coder harness": "", "HumanEval /164": "163", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "reference", "Date": "2026-10-01; 2026-11-10" }, { "Model": "Qwen2.5-Coder-14B-Instruct (stock), zero-shot", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "216-task synthetic holdout (pre-rebuild harness): 3.2% of 216 (7/216)", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "superseded by Qwen3.8-27B line", "Date": "2026-09-05 (and earlier)" }, { "Model": "Qwen2.5-Coder-14B-Instruct (stock) + 4 few-shot protocol exemplars", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "216-task synthetic holdout (pre-rebuild harness): 21.3% of 216 (46/216)", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "40 of 46 passes ended at the step limit; real bench 2/38 RETRACTED", "Verdict": "superseded by Qwen3.8-27B line", "Date": "2026-09-05 (and earlier)" }, { "Model": "House 14B fine-tune v2 (14B, theology v11 base, 3,978-example corpus)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "216-task synthetic holdout (pre-rebuild harness): 28.2%", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "superseded by Qwen3.8-27B line", "Date": "2026-09-05 (and earlier)" }, { "Model": "House 14B fine-tune v3 (14B, theology v11 base, 4,721 examples)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "216-task synthetic holdout (pre-rebuild harness): 27.3% old harness / 33.8% fixed harness", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "superseded by Qwen3.8-27B line", "Date": "2026-09-05 (and earlier)" }, { "Model": "House 14B fine-tune v4 (Qwen2.5-Coder-14B base, 4,721 examples)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "216-task synthetic holdout (pre-rebuild harness): 39.4% / 44.0% (box variance); traces and both 95/216", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "base swap theology->code +5.32 pp pooled, McNemar p=0.0081", "Verdict": "superseded by Qwen3.8-27B line", "Date": "2026-09-05 (and earlier)" }, { "Model": "House 14B fine-tune v5 (Qwen2.5-Coder-14B base, 4,968 examples incl. 247 on-policy)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "216-task synthetic holdout (pre-rebuild harness): 43.5%", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "superseded by Qwen3.8-27B line", "Date": "2026-09-05 (and earlier)" }, { "Model": "Ripple-65B-Preview (arch 'ricursive'; base model not recorded)", "Uploader / repo": "'-AD-' build (repo not recorded), A100 server", "Quant": "IQ3_XXS", "File size": "27G", "Hardware / loader": "RTX PRO 6000 per session log (yaml points at), Ripple-patched llama.cpp", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "A, net-off: VOID, not scored. 61 of 100 attempted, 0 non-empty patches (max_tokens 4096 + 16k/slot config bug; 306 ContextWindowExceeded log hits)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "DELETED 2026-11-08 (we: 'ripple is trash', 'burn all these models'); do not re-propose", "Date": "2026-11-06" }, { "Model": "Ripple-65B-Preview (arch 'ricursive'; base model not recorded)", "Uploader / repo": "(build not recorded)", "Quant": "IQ3 (; exact type not recorded)", "File size": "", "Hardware / loader": "RTX PRO 6000, Ripple llama.cpp built for Blackwell", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "A, net-off: VOID, 21 of 100 attempted, 0 non-empty patches (out_ripple_iq3_clean)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "DELETED 2026-11-08 (we: 'ripple is trash', 'burn all these models'); do not re-propose", "Date": "2026-11-06" }, { "Model": "Ripple-65B-Preview (arch 'ricursive'; base model not recorded)", "Uploader / repo": "'-AD-' build", "Quant": "IQ4_XS", "File size": "37G", "Hardware / loader": "A100 server 2x A100 40 GB, run in parallel with other quants (both GPUs 99% full)", "Gen tok/s": "4.1", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "A, net-off: 0 predictions written (every task timed out); not scored", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "DELETED 2026-11-08 (we: 'ripple is trash', 'burn all these models'); do not re-propose", "Date": "2026-11-06" }, { "Model": "Ripple-65B-Preview (arch 'ricursive'; base model not recorded)", "Uploader / repo": "'-AD-' build", "Quant": "Q6_K", "File size": "54G (42G+12G)", "Hardware / loader": "A100 server 2x A100 40 GB alone, Ripple llama.cpp (build-a100)", "Gen tok/s": "16.9 (17.2 in server log)", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "A, net-off: VOID. out_ripple_q6_100: 20 attempted, 0 non-empty. Report 'out_ripple_q6_partial': 60 submitted, 58 empty, 2 resolved (2/2 real patches). Fixed-config probe django-11299: 1/1 resolved", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "DELETED 2026-11-08 (we: 'ripple is trash', 'burn all these models'); do not re-propose", "Date": "2026-11-06" }, { "Model": "Ripple-65B-Preview (arch 'ricursive'; base model not recorded)", "Uploader / repo": "non-AD build", "Quant": "Q4_K_M", "File size": "37G", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "never tested ( exists)", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "DELETED 2026-11-08 (we: 'ripple is trash', 'burn all these models'); do not re-propose", "Date": "2026-11-06" }, { "Model": "Ripple-65B-Preview (arch 'ricursive'; base model not recorded)", "Uploader / repo": "non-AD build", "Quant": "Q8_0", "File size": "65G (42G+23G)", "Hardware / loader": "A100 server 2x A100 40 GB, Ripple llama.cpp, -c 131072 --parallel 2 (65,536/slot), max_tokens 12000", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "A, net-off, NOT on the board: run died at 52/100 (context overflow). 52 submitted, 29 non-empty, 26 resolved, 3 unresolved, 23 empty (26/29 = ~90% of real patches). Earlier partial 23/26", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "DELETED 2026-11-08 (we: 'ripple is trash', 'burn all these models'); do not re-propose", "Date": "2026-11-07" }, { "Model": "Qwen2.5-Coder-14B-Instruct abliterated (MSJXMG)", "Uploader / repo": "MSJXMG", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "216-task shootout, 2026-09-07, run with network ON (not trusted): 21.8%, refusals 0%", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "not chosen", "Date": "2026-09-07" }, { "Model": "Qwen2.5-Coder-14B-Instruct abliterated (aesdi)", "Uploader / repo": "aesdi", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "216-task shootout, 2026-09-07, run with network ON (not trusted): 21.3%, refusals 12%", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "not chosen", "Date": "2026-09-07" }, { "Model": "Qwen2.5-Coder-14B-Instruct abliterated (criscar)", "Uploader / repo": "criscar", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "216-task shootout, 2026-09-07, run with network ON (not trusted): 20.8%, refusals 0%", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "not chosen", "Date": "2026-09-07" }, { "Model": "Qwen2.5-Coder-14B-Instruct (stock) in the 2026-09-07 shootout", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "216-task shootout, 2026-09-07, run with network ON (not trusted): 19.4%, refusals 100%", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "Josiefied and sci4ai 14B variants: preflight-fail", "Verdict": "", "Date": "2026-09-07" }, { "Model": "Kimi K2.7 Code abliterated, in-house bake 'kimi-abl-Q4_K_M'", "Uploader / repo": "", "Quant": "Q4_K_M", "File size": "578 GB (620.5 GB per session log)", "Hardware / loader": "Mac Studio Metal", "Gen tok/s": "~5 (session log: 'big bakes ran ~6 tok/s on Metal')", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "shelved (Q3_K_XL 494 GB and Q3C417 365 GB bakes never run)", "Date": "2026-11" }, { "Model": "Qwen2.5-14B-Instruct-1M abliterated", "Uploader / repo": "", "Quant": "Q4_K_M (q4_0 KV)", "File size": "", "Hardware / loader": "A100 80 GB, llama.cpp -c 1048576", "Gen tok/s": "", "Prompt tok/s": "573 (prefill)", "Refusals /8": "not run", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "1M-context needle test: PASS ~250k (252k, 440 s), FAIL ~500k (515k)", "Verdict": "not chosen", "Date": "2026-09-14" }, { "Model": "Qwen2.5-7B-Instruct-1M abliterated", "Uploader / repo": "", "Quant": "Q8_0 (q8_0 KV), llama.cpp", "File size": "", "Hardware / loader": "A100 80 GB, llama.cpp -c 1048576", "Gen tok/s": "", "Prompt tok/s": "1,639 (prefill)", "Refusals /8": "not run (llama.cpp round)", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "needle PASS 252k (154 s), FAIL 515k", "Verdict": "see vLLM row", "Date": "2026-09-14" }, { "Model": "Qwen2.5-7B-Instruct-1M abliterated", "Uploader / repo": "", "Quant": "bf16 dequantised from Q8_0, vLLM with Dual Chunk Attention", "File size": "", "Hardware / loader": "A100 80 GB (round 11, 700k window); RTX PRO 6000 96 GB (round 15, 965k window, 56 GB KV, last_q=32)", "Gen tok/s": "~39 (RTX PRO 6000, 300 tokens in 7.7 s)", "Prompt tok/s": "~3.0k at 958k (RTX PRO 6000)", "Refusals /8": "0/8", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "A100: 3 of 4 needles PASS at 552k. RTX: needles PASS at 552k, 785k, 958k (15/50/85% depth). Rounds 2-7 void (vLLM build lacked DCA)", "Verdict": "GO; shipped as our house build", "Date": "2026-09-14 .. later rounds" }, { "Model": "Nemotron-8B-UltraLong-1M abliterated", "Uploader / repo": "", "Quant": "Q8_0 (q4_0 KV)", "File size": "", "Hardware / loader": "A100 80 GB, llama.cpp -c 1048576", "Gen tok/s": "", "Prompt tok/s": "1,306 (prefill)", "Refusals /8": "0/8", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "needle FAIL at 201k", "Verdict": "rejected", "Date": "2026-09-14" }, { "Model": "Qwen3-Coder-30B-A3B-Instruct abliterated (1M YaRN context)", "Uploader / repo": "huihui-ai", "Quant": "Q4_K_M", "File size": "", "Hardware / loader": "A100 80 GB, llama.cpp", "Gen tok/s": "", "Prompt tok/s": "1,169 at 219k; 350 at 757k", "Refusals /8": "0/8", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "needle PASS 219,325 tok and 250,041 tok (native 256k); FAIL at 411k and 757k", "Verdict": "", "Date": "1M-context study (date per source)" }, { "Model": "Qwen3-4B abliterated", "Uploader / repo": "mlabonne/Qwen3-4B-abliterated", "Quant": "MLC q4f32_1", "File size": "2.28 GB, 74 shards", "Hardware / loader": "WebGPU in Chrome (WebLLM) on RTX 5070 desktop; our Intel laptop", "Gen tok/s": "median 34.2 (P10 25.5-31.9) in-page; 41.1-46.6 via our site; 2.4 on Intel laptop (hard-froze it)", "Prompt tok/s": "first token 290 ms", "Refusals /8": "0/8 (in-page probe PASS 8 valid / 0 refused / 0 errors)", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "KL vs f16 (shipped q4f32_1): 0.1201, top-1 83.03%; q4f32_1-awq 0.0921/84.90%; q4f16_1-awq 0.0903/84.64%. 12-question check: 11/12 (n=12, not significant)", "Verdict": "shipped edge model", "Date": "2026-09 (per)" }, { "Model": "Qwen3-4B abliterated (GGUF quant matrix)", "Uploader / repo": "mlabonne (GGUF by bartowski for Q4_K_M)", "Quant": "Q8_0 4.28 GB / Q4_K_M 2.50 GB / Q4_K_M+imatrix 2.50 GB / IQ4_XS+imatrix 2.27 GB / Q4_0 2.37 GB", "File size": "", "Hardware / loader": "ollama on server T4 (gate); quality via llama.cpp KL vs f16 (PPL 25.72)", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "0/8 (Q4_K_M, valid 8)", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "mean KL / top-1: Q8_0 0.012763 / 93.231%; Q4_K_M 0.046079 / 88.066%; Q4_K_M+imat 0.035986 / 89.406%; IQ4_XS+imat 0.041191 / 88.563%; Q4_0 0.070062 / 85.635%. Erratum: Q8 '5x fragile' reading withdrawn", "Verdict": "reference for the edge build", "Date": "2026-09" }, { "Model": "Qwen3-1.7B abliterated (edge)", "Uploader / repo": "", "Quant": "", "File size": "939 MB", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "12-question check: 8/12 (n=12)", "Verdict": "", "Date": "2026-09" }, { "Model": "Qwen3-0.6B abliterated (edge)", "Uploader / repo": "", "Quant": "MLC q4f16_1 / q4f32_1", "File size": "336 MB", "Hardware / loader": "WebGPU on RTX 5070 desktop", "Gen tok/s": "~15", "Prompt tok/s": "", "Refusals /8": "0/8 (8 valid, none refused)", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "12-question check: 7/12 (n=12)", "Verdict": "", "Date": "2026-09" }, { "Model": "Qwen3-VL-8B abliterated", "Uploader / repo": "ollama huihui_ai/qwen3-vl-abliterated:8b", "Quant": "Q4_K_M", "File size": "6.1 GB", "Hardware / loader": "ollama on a T4", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "0/8 (2026-09-16); later gate run valid=2 errors=6 refusals=0 = FAIL (2026-09-30)", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "vision model trial", "Date": "2026-09-16 / 2026-09-30" }, { "Model": "our house build 14B v13 (SFT on Huihui-Qwen3-14B-abliterated-v2)", "Uploader / repo": "", "Quant": "BF16", "File size": "28-29.5 GB", "Hardware / loader": "RTX PRO 6000 llama.cpp (prod, then RTX PRO 6000 box; vLLM on A100 (load test)", "Gen tok/s": "52.4 single user, 305 total at 32 users (vLLM A100 load test)", "Prompt tok/s": "", "Refusals /8": "2/8 and 2/8 + 1 error (2026-09-30, FAIL); 7/8 at build time (quoted)", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "LIVE as our house build (theology model)", "Date": "2026-09-30; 2026-10-18; 2026-11-13" }, { "Model": "our house build 14B v12 (SFT on Huihui-Qwen3-14B-abliterated-v2)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "", "Gen tok/s": "", "Prompt tok/s": "", "Refusals /8": "1/8 at build time (quoted)", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "Failed our house build Auto router screener test (2026-11-06)", "Verdict": "", "Date": "2026-09; 2026-11-06" }, { "Model": "Qwen3.8-27B abliterated (our house build 3.0) — load test", "Uploader / repo": "huihui-ai", "Quant": "Q4 GGUF", "File size": "", "Hardware / loader": "A100, llama.cpp, 2 slots", "Gen tok/s": "33.9 single user, ~70 total ceiling", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "", "Date": "(per source)" }, { "Model": "Qwen3.8-Flash-Next abliterated (V100 run)", "Uploader / repo": "huihui-ai", "Quant": "UD-Q4_K_XL GGUF", "File size": "", "Hardware / loader": "V100 server 8x V100, llama.cpp", "Gen tok/s": "~4 per stream", "Prompt tok/s": "", "Refusals /8": "1/8 (slurs probe)", "SWE-bench Verified 100": "", "216-task coder harness": "abandoned at 93/216 (partial: Flash 50 vs 27B 64 on same tasks)", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "abandoned", "Date": "2026-10-04" }, { "Model": "Qwen2.5 14B (ollama qwen2.5:14b)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "ollama on T4s", "Gen tok/s": "~15 per card (~45 total)", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "", "Date": "2026-08-25" }, { "Model": "Llama 3.1 8B (ollama llama3.1:8b)", "Uploader / repo": "", "Quant": "", "File size": "", "Hardware / loader": "ollama on T4s", "Gen tok/s": "30.8 one card; 71.4 total", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "", "Date": "2026-08-25" }, { "Model": "Qwen3-14B abliterated", "Uploader / repo": "ollama richardyoung/qwen3-14b-abliterated", "Quant": "q4_K_M", "File size": "", "Hardware / loader": "ollama on T4s", "Gen tok/s": "28.7 total", "Prompt tok/s": "", "Refusals /8": "", "SWE-bench Verified 100": "", "216-task coder harness": "", "HumanEval /164": "", "HumanEval+ / MBPP+ / LiveCodeBench": "", "Other results": "", "Verdict": "", "Date": "2026-08-25" } ]