Leaderboard
The holistic, weighted roll-up that sits above the gauntlet. Overall blends capability 40% · efficiency 35% · reliability 25%. Capability is a domain-weighted solve-rate; efficiency is throughput; reliability is clean-completion. Models below the 3-domain coverage gate sit in provisional.
🏆 Overall
holistic weighted score · 22 rankedbest across all harnesses — capability ceilingcompare harness impact →| # | Model | Overall | Cap | Eff | Rel | Dom | Last run |
|---|---|---|---|---|---|---|---|
| 1 | qwen3.6-35b-a3b cloud Alibaba · 35B-A3B | 99.3 | 98.3 | 100 | 100 | 4/5 | 29d |
| 2 | gpt-oss-20b · MXFP4MXFP4 OpenAI · 20B | 89.9 | 99.2 | 72 | 100 | 4/5 | 29d |
| 3 | gpt-oss-120b · MXFP4MXFP4 OpenAI | 81.6 | 100 | 47.5 | 100 | 4/5 | 43d |
| 4 | qwen3.5-4b · Q8Q8 Alibaba · 4B | 78.5 | 95.4 | 44.8 | 98.5 | 4/5 | 28d |
| 5 | qwen3.5-2b · Q8Q8 Alibaba | 77.6 | 76.7 | 62.8 | 100 | 4/5 | 28d |
| 6 | qwen3.6-35b-a3b · BF16BF16 Alibaba · 35B-A3B | 75.8 | 97.1 | 34.2 | 100 | 4/5 | 28d |
| 7 | qwen3.5-0.8b · Q8Q8 Alibaba | 74.5 | 52.1 | 81.9 | 100 | 4/5 | 28d |
| 8 | qwen3-coder-30b · Q8Q8 Alibaba · 30B-A3B | 73.5 | 87.9 | 38.2 | 100 | 4/5 | 29d |
| 9 | claude-sonnet-4.6 cloud | 71.8 | 98.3 | 21.5 | 100 | 4/5 | 24d |
| 10 | granite-4.1-8b · Q8Q8 IBM | 67.9 | 89.2 | 20.6 | 100 | 4/5 | 43d |
| 11 | qwen3.5-122b-a10b-cluster | 67.3 | 92.9 | 16.7 | 97.3 | 4/5 | 31d |
| 12 | granite-4.1-30b · Q8Q8 IBM · 30B | 66.6 | 98.3 | 6.4 | 100 | 4/5 | 28d |
| 13 | qwen3.6-27b Alibaba | 66.4 | 98.3 | 8.1 | 96.8 | 4/5 | 25d |
| 14 | qwen3.6-35b-a3b · Q8Q8 Alibaba · 35B-A3B | 65.7 | 56.4 | 53.9 | 97.1 | 5/5 | 25d |
| 15 | gemini-2.5-flash-lite cloud | 65.7 | 79.6 | 25.3 | 100 | 4/5 | 29d |
| 16 | llama-3.3-70b · Q8Q8 Meta | 65.2 | 98.1 | 2.9 | 100 | 4/5 | 42d |
| 17 | mistral-small-3.1-24b-instruct-2503 · Q8Q8 Mistral AI · 24B | 65.1 | 92.9 | 8.4 | 100 | 4/5 | 43d |
| 18 | devstral-small-2-2512 · Q8Q8 Mistral AI · 24B | 58.6 | 77.2 | 8.4 | 99.2 | 5/5 | 42d |
| 19 | phi-4-reasoning-plus · Q8Q8 Microsoft | 56.8 | 69.2 | 14.8 | 95.6 | 4/5 | 27d |
| 20 | qwen3-next-80b · Q8Q8 Alibaba · 80B-A3B | 56.2 | 51.7 | 33.5 | 95.3 | 5/5 | 28d |
| 21 | gemma-4-26b-a4b-it · Q8Q8 Google · 26B-A4B | 52.9 | 48.2 | 27.2 | 96.6 | 5/5 | 29d |
| 22 | deepseek-r1-distill-qwen-32b DeepSeek | 37.8 | 31.2 | 5.1 | 94.2 | 4/5 | 27d |
Provisional — under 3-domain coverage, not yet ranked
| # | Model | Overall | Cap | Eff | Rel | Dom | Last run |
|---|---|---|---|---|---|---|---|
| · | phi-4 | 100 | 100 | — | 100 | 1/5 | never |
| · | deepseek-r1-32b | 83.8 | 80 | — | 90 | 1/5 | never |
| · | hermes-4.3-36b | 11.2 | 18.2 | — | 0 | 1/5 | never |
🌱 Up & comers
new in last 14dNo new arrivals clearing the bar in this window.