ChessBench - Coherence: leaderboard
ChessBench rule-compliance score: the fraction of attempted moves that were legal, weighted by the fraction of games played fully legally.
Metric: Coherence (%). Source: chessbench.ai. Saturation forecast: Around February 2027. 72 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 100 | #33 |
| 2 | Muse Wonder | 100 | #18 |
| 3 | GPT-5.1 | 97.9 | #145 |
| 4 | GPT-6 Astra | 97 | #6 |
| 5 | GPT-5.4 | 95.3 | #83 |
| 6 | Gemini 3.8 Flash | 94.6 | #30 |
| 7 | Gemini 4 Argon | 92.6 | #9 |
| 8 | GPT-5 | 91.4 | #105 |
| 9 | Muse Spark 1.3 | 91.2 | #31 |
| 10 | Muse Spark 1.2 | 89.7 | #69 |
| 11 | GPT-5.6 Sol | 89.1 | #22 |
| 12 | GPT-5 Nano | 88.6 | #439 |
| 13 | Muse Spark 1.1 | 87.6 | #55 |
| 14 | GPT-6.1 Sol | 87.5 | #7 |
| 15 | Gemini 3 Flash (Preview) | 87.3 | #90 |
Interactive version: theaggregate.ai/benchmark?slug=chessbench-coherence · How It Works · Data refreshed daily, snapshot 2026-11-05.