ChessBench Coherence
Games · 2026-07-28
ChessBench (chessbench.ai) has language models play full chess games against reference opponents and scores the moves; it is unrelated to the chess-bench.com puzzle benchmark carried as `chessbench`. Coherence is the rate of rule-compliant play, defined by the leaderboard as move_coherence * max(0.01, game_coherence), where move_coherence is the fraction of attempted moves that were legal and game_coherence the fraction of fully-legal games. A product of two rates rather than a proportion of scored items, but bounded on [0, 1] by construction and reported here in percentage points.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.5 | 100.0 |
| GPT-5.1 | 97.9 |
| Muse Spark 1.2 | 97.9 |
| GPT-6 Astra | 97.0 |
| GPT-5.4 | 95.3 |
| Gemini 3.8 Flash | 94.5 |
| GPT-5 | 92.8 |
| Gemini 4 Argon | 92.6 |
| Muse Spark 1.3 | 91.2 |
| GPT-5.6 Sol | 89.1 |
| GPT-5 Nano | 88.6 |
| Muse Spark 1.1 | 88.6 |
| GPT-6 Sol | 88.2 |
| GPT-6.1 Sol | 87.5 |
| Gemini 3.6 Flash | 87.0 |
| Gemini 3.1 Pro Preview | 85.9 |
| o3 | 84.7 |
| Gemini 3.7 Flash | 84.0 |
| Opus 4.6 | 83.9 |
| GPT-5.6 Luna | 83.1 |
| Claude Fable 5 | 82.6 |
| Claude Haiku 5.5 | 80.7 |
| Claude Opus 5.5 | 80.1 |
| Opus 4.8 | 79.7 |
| Sonnet 4.6 | 79.1 |
| GPT-5.2 | 79.1 |
| Gemini 3.5 Flash | 78.1 |
| GPT-5.4 Nano | 77.8 |
| Gemini 3 Flash Preview | 76.3 |
| GPT-6 Luna | 75.4 |
| GPT-5.4 Mini | 75.2 |
| O4 Mini | 73.0 |
| GPT-5.6 Terra | 69.7 |
| Grok 4.6 | 69.2 |
| Sonnet 5 | 67.8 |
| GPT-5 Mini | 65.7 |
| Grok 4.3 | 65.0 |
| Claude Sonnet 5.5 | 64.6 |
| Claude Opus 5 | 63.8 |
| Grok 4.20 | 62.6 |