Benchmarks › Chess Puzzles
74top score · GPT-6 Astra
68models ranked
11open-weights models
Accuracyhigher is better
Oct 20, 2026last updated
How the test works
1
Task
100 chess positions
- What the model doesFind strong moves in difficult chess positions
- Kinds of tasksBest-next-move selection
2
Run
Inspect AI qa_agent (v2)
- Token budget1,000,000 output tokens; compaction at 90% of the model context window
3
Score
Accuracy
- Graded byDeterministic normalized exact match to the answer key (v2)
- Better isHigher
Sources: epoch.ai, github.com
Chess Puzzles leaderboardtop 10 of 68
Full leaderboard with filters →| Model | Chess Puzzles | Per task | Price | Speed | Supports | Context |
|---|---|---|---|---|---|---|
74max | $3.26 | $10 / $50 | 69 | 1.05M | ||
68max | $0.72 | $2 / $10 | 52 | 1.05M | ||
67max | $5.98 | $4 / $20 | 96 | 1M | ||
63xhigh | $2.63 | $5 / $30 | 113 | 1.05M | ||
62xhigh | — | $30 / $180 | — | 1.05M | ||
59max | $5.46 | $2 / $10 | 137 | 1M | ||
59high | $1.99 | $2 / $10 | 84 | 1M | ||
59max | $1.99 | $4 / $20 | 114 | 1.05M | ||
59 | — | $30 / $180 | — | 1.05M | ||
58high | $1.24 | $0.75 / $3.75 | 149 | 1M |
Chess Puzzles questions
What is Chess Puzzles?›
Find strong moves in difficult chess positions. It has 100 chess positions. Created by Epoch AI.
Which model is best on Chess Puzzles?›
GPT-6 Astra leads Chess Puzzles with 74, ahead of GPT-6.1 Sol (68) and Claude Opus 5.5 (67).
What is the best open-weights model on Chess Puzzles?›
DeepSeek V4 Pro 0813 with 46, then Kimi K3 (37).
How is Chess Puzzles scored?›
Accuracy; deterministic normalized exact match to the answer key (v2).