Benchmarks › Mystery Game Puzzles
84top score · GPT-6 Astra
64models ranked
15open-weights models
Accuracyhigher is better
Oct 20, 2026last updated
How the test works
1
Task
100 game positions
- What the model doesSolve puzzles from an undisclosed game variant
- Kinds of tasksBest-next-move selection
2
Run
Minimal agent scaffold with a submission tool
- Tools allowedMove submission
3
Score
Accuracy
- Graded byNormalize submitted move and compare to answer key; invalid/unanswered moves get no credit
- Better isHigher
- Test setGame identity, prompts, examples and model transcripts withheld
Sources: epoch.ai
Mystery Game Puzzles leaderboardtop 10 of 64
Full leaderboard with filters →| Model | Mystery Game Puzzles | Per task | Price | Speed | Supports | Context |
|---|---|---|---|---|---|---|
84max | $3.26 | $10 / $50 | 69 | 1.05M | ||
80max | $0.72 | $2 / $10 | 52 | 1.05M | ||
71max | $5.98 | $4 / $20 | 96 | 1M | ||
65max | $5.46 | $2 / $10 | 137 | 1M | ||
64high | $1.99 | $2 / $10 | 84 | 1M | ||
59max | $5.86 | $5 / $25 | 55 | 1M | ||
58max | $7.63 | $10 / $50 | 69 | 1M | ||
58max | $1.99 | $4 / $20 | 114 | 1.05M | ||
56xhigh | $2.63 | $5 / $30 | 113 | 1.05M | ||
56max | $1.04 | $2 / $10 | 128 | 1.05M |
Mystery Game Puzzles questions
What is Mystery Game Puzzles?›
Solve puzzles from an undisclosed game variant. It has 100 game positions. Created by Epoch AI.
Which model is best on Mystery Game Puzzles?›
GPT-6 Astra leads Mystery Game Puzzles with 84, ahead of GPT-6.1 Sol (80) and Claude Opus 5.5 (71).
What is the best open-weights model on Mystery Game Puzzles?›
DeepSeek V4 Pro 0813 with 43, then GLM 5.3 (33).
How is Mystery Game Puzzles scored?›
Accuracy; normalize submitted move and compare to answer key; invalid/unanswered moves get no credit.