AI Chess Leaderboard (Reasoning): leaderboard
The full-information mode of the dubesor AI chess project. Each turn supplies the position in FEN, the move history in SAN and the legal moves, and the model returns reasoning and one move as JSON; games average about 51 turns. Ratings start from Stockfish 17.1 accuracy over a model's first ten games, then update by standard Elo on games between different models, so they are relative to this pool; higher is better.
Metric: Elo. Source: dubesor.de. Saturation forecast: Around September 2027. 345 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 1940 | #58 |
| 2 | Gemini 3.7 Flash | 1851 | #23 |
| 3 | Gemini 3 Pro (Preview) | 1850 | #66 |
| 4 | Gemini 3.8 Flash | 1847 | #21 |
| 5 | Qwen 3 Max (Thinking) | 1800 | #208 (Qwen 3 Max) |
| 6 | Step 5 Preview | 1800 | #44 |
| 7 | Ling 3.1 Flash | 1800 | #71 |
| 8 | GPT-5 Codex | 1777 | #127 |
| 9 | Gemini 3.6 Flash | 1753 | #51 |
| 10 | Claude Opus 5.5 | 1749 | #3 |
| 11 | GPT-5.1 Codex | 1743 | #159 |
| 12 | GPT-6 Astra | 1736 | #6 |
| 13 | Gemini 3.5 Flash | 1721 | #60 |
| 14 | GPT-5.5 | 1668 | #31 |
| 15 | Gemini 4 Argon | 1638 | #8 |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=ai-chess-leaderboard-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.