Chess Puzzles (Epoch AI): leaderboard
One item is a chess position supplied as a FEN string, and the model must finish its reply with the single best next move written as a start square and end square. Epoch AI generated the 100 positions with an engine and treats Stockfish's choice as the key, pulling the move out of the response with an extractor model and checking it for an exact match. Higher is better.
Metric: Accuracy (%). Source: epoch.ai. Saturation forecast: Around June 2027. 153 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 74 | #6 (GPT-6 Astra) |
| 2 | GPT-6.1 Sol (Max) | 68 | #7 (GPT-6.1 Sol) |
| 3 | Claude Opus 5.5 (Max) | 67 | #3 (Claude Opus 5.5) |
| 4 | Gemini 3.7 Flash (High) | 66 | #23 (Gemini 3.7 Flash) |
| 5 | GPT-5.5 (xHigh) | 63 | #31 (GPT-5.5) |
| 6 | GPT-5.5 Pro (xHigh) | 62 | #17 (GPT-5.5 Pro) |
| 7 | GPT-5.6 Sol (Max) | 59 | #19 (GPT-5.6 Sol) |
| 8 | Gemini 4 Argon (High) | 59 | #8 (Gemini 4 Argon) |
| 9 | Claude Sonnet 5.5 (Max) | 59 | #11 (Claude Sonnet 5.5) |
| 10 | Gemini 3.8 Flash (High) | 58 | #21 (Gemini 3.8 Flash) |
| 11 | GPT-6 Sol (Max) | 56 | #14 (GPT-6 Sol) |
| 12 | Claude Haiku 5.5 (Max) | 55 | #40 (Claude Haiku 5.5) |
| 13 | Claude Opus 4.8 (Max) | 52 | #37 (Claude Opus 4.8) |
| 14 | Gemini 3.1 Pro (Preview) (High) | 51 | #58 (Gemini 3.1 Pro (Preview)) |
| 15 | DeepSeek V4.1 Flash (Max) | 48 | #36 (DeepSeek V4.1 Flash) |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=chess-puzzles-epoch-ai · How It Works · Data refreshed daily, snapshot 2026-10-20.