| Model | N | Optimal | Avg regret | J / decision | Efficiency |
|---|---|---|---|---|---|
| gemini-3.1-pro-preview | 62 | 100.0% | 0.000 | 12,238 | |
| claude-fable-5 | 49 | 100.0% | 0.000 | 48,156 | |
| claude-opus-5 | 39 | 100.0% | 0.000 | 21,202 | |
| gemini-3.7-flash | 238 | 100.0% | 0.000 | 1,433 | |
| gemma-4-31b-it | 83 | 100.0% | 0.000 | 1,114 | |
| grok-4.6 | 246 | 99.6% | 0.004 | 10,491 | |
| gpt-5.6-sol | 132 | 98.5% | 0.015 | 10,960 | |
| qwen3.8-27b | 70 | 97.1% | 0.029 | 6,864 | |
| claude-sonnet-5 | 352 | 97.4% | 0.026 | 8,932 | |
| gpt-5.6-terra | 133 | 94.0% | 0.060 | 6,526 | |
| gpt-oss-120b | 150 | 90.0% | 0.100 | 377 | |
| glm-5.3 | 44 | 97.7% | 0.023 | 3,455 | |
| muse-spark-1.2 | 66 | 92.4% | 0.076 | 7,477 | |
| gpt-5.6-luna | 280 | 90.0% | 0.104 | 943 | |
| gemini-3.5-flash-lite | 101 | 81.2% | 0.188 | 899 | |
| claude-haiku-4.5 | 83 | 73.5% | 0.301 | 3,307 | |
| inkling | 28 | 60.7% | 0.464 | 2,129 |
Efficiency Rating is the fraction of a model’s metered capacity that bought an optimal move, bounded to [0,1] per match. It says nothing about how often a model won — browse every scored decision for that, or rescore a match to see one worked end to end.