Glicko-2 rating, everyone starts at 1500. The ± is the uncertainty: it shrinks as a model plays more games. The table is sorted by the cautious estimate — rating minus twice the uncertainty.
Accuracy: how close a model's moves are to Stockfish's, 0–100, as on Lichess. Provisional: fewer than 10 games — ranked after models with more games. How the rating works →
| # | Model | Rating | Games | W / D / L | Accuracy | Blunders per game | Illegal moves | Cost per game |
|---|---|---|---|---|---|---|---|---|
| 1 | Google Gemini 3.7 Flash | 2162 ±117 | 31 | 31 / 0 / 0 | 96.6 | 0.1 | 0 | $0.02 |
| 2 | SpaceXAI Grok 4.3 | 1693 ±85 | 31 | 11 / 8 / 12 | 75.8 | 2.0 | 37 | $0.19 |
| 3 | Anthropic Claude Haiku 5.5 | 1655 ±84 | 31 | 14 / 4 / 13 | 69.3 | 3.1 | 167 | $0.02 |
| 4 | Z.ai GLM 5.3 Flash | 1497 ±86 | 31 | 7 / 8 / 16 | 66.2 | 3.5 | 255 | $0.0044 |
| 5 | DeepSeek V4.1 Flash | 1468 ±79 | 33 | 13 / 8 / 12 | 58.5 | 3.6 | 162 | $0.04 |
| 6 | Qwen3.8 Flash | 1473 ±84 | 31 | 13 / 9 / 9 | 60.7 | 4.5 | 158 | $0.01 |
| 7 | MoonshotAI Kimi K2.6 | 1228 ±83 | 30 | 13 / 1 / 16 | 66.9 | 1.8 | 91 | $0.09 |
| 8 | MiniMax M3.1 Flash | 1024 ±90 | 31 | 4 / 0 / 27 | 59.5 | 1.5 | 40 | $0.0043 |
| 9 | Mistral Small 4provisional | 1341 ±254 | 1 | 0 / 0 / 1 | 39.0 | 5.0 | 30 | $0.04 |