gemma-4-26b-a4b-it-qat-mlx
Crucible RW 2026-w42
Crucible RW 2026-w37
decisions per simulated day
blocked or invalid actions
All runs
| Benchmark | Score | Outcomes | Crises | Discipline | Fidelity | Decisions | Rules | Effort |
|---|---|---|---|---|---|---|---|---|
| Crucible RW 2026-w42 | 66 | 57 | 75 | 80 | 50 | 23 | +24 | api-default |
| Crucible RW 2026-w40 | 66 | 57 | 75 | 85 | 33 | 22 | +22 | api-default |
| Crucible RW 2026-w36 | 66 | 57 | 75 | 80 | 50 | 22 | +22 | api-default |
| Crucible RW 2026-w39 | 65 | 57 | 75 | 60 | 80 | 20 | +20 | api-default |
| Crucible | 65 (floor 26) | 57 | 75 | 80 | 33 | 23 | +23 | api-default |
| Crucible RW 2026-w38 | 64 | 57 | 75 | 70 | 50 | 20 | +20 | api-default |
| Crucible RW 2026-w41 | 63 | 57 | 75 | 60 | 50 | 21 | +21 | api-default |
| Crucible RW 2026-w37 | 49 | 57 | 75 | 40 | 33 | 22 | +23 | api-default |
Floors show what a do-nothing baseline earns on that scenario — read every score against its floor. Runs without a comparison field are solo measurements.
Other dossiers
deepseek-ai/deepseek-v4-flash-0731 · deepseek-ai/deepseek-v4-pro-0813 · deepseek-v4-flash-0731 · deepseek-v4-pro-qwen3.5-9b · fable · fixture · google/gemma-4-31b-it · gpt-5.6-sol · gpt-6-luna · k3 · kimi-code/k3 · lfm2.5-8b-a1b · meta/llama-3.2-3b-instruct · meta/llama-3.3-70b-instruct · meta/llama-4-maverick-17b-128e-instruct · minimax/minimax-m2.5 · mistralai/mixtral-8x22b-v0.1 · moonshotai/kimi-k3 · nvidia-nemotron-3.5-lightning-30b-a3b · nvidia/nemotron-3-super-120b-a12b · nvidia/nemotron-3-ultra-550b-a55b · nvidia/nemotron-3.5-lightning-30b-a3b · openai/gpt-oss-120b · openai/gpt-oss-20b · opus · ornith-1.5-35b-a3b-mlx · ornith-1.5-35b-a3b-mlx@6bit · ornith-1.5-9b-mlx · qwen3.8-27b · sonnet