Epoch AI - Mystery Game Puzzles: leaderboard
Epoch AI Mystery Game Puzzles evaluates game reasoning on 100 mid-game positions from a deliberately undisclosed board game, each filtered to have a single best next move; the game's identity is withheld to limit benchmark-specific training. Models read the state as text and submit one move via a tool call, normalized to the game's notation and checked against an answer key — illegal or unextractable moves score zero. Score is accuracy (%). Structured like Epoch AI's Chess Puzzles board but on an unknown game.
Metric: Score. Source: epoch.ai. Saturation forecast: Around December 2027. 143 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 84 | #6 (GPT-6 Astra) |
| 2 | GPT-6.1 Sol (Max) | 80 | #7 (GPT-6.1 Sol) |
| 3 | GPT-6.1 Terra (Max) | 74 | #15 (GPT-6.1 Terra) |
| 4 | Claude Opus 5.5 (Max) | 71 | #3 (Claude Opus 5.5) |
| 5 | Claude Sonnet 5.5 (Max) | 65 | #14 (Claude Sonnet 5.5) |
| 6 | Gemini 4 Argon (High) | 64 | #9 (Gemini 4 Argon) |
| 7 | Claude Opus 5 (Max) | 59 | #16 (Claude Opus 5) |
| 8 | GPT-5.6 Sol (Max) | 58 | #22 (GPT-5.6 Sol) |
| 9 | Claude Fable 5.1 (Max) | 58 | #10 (Claude Fable 5.1) |
| 10 | GPT-5.5 (xHigh) | 56 | #33 (GPT-5.5) |
| 11 | GPT-6 Sol (Max) | 56 | #19 (GPT-6 Sol) |
| 12 | GPT-5.5 (High) | 52 | #33 (GPT-5.5) |
| 13 | Claude Fable 5 (Max) | 52 | #20 (Claude Fable 5) |
| 14 | Muse Wonder (Max) | 51 | #18 (Muse Wonder) |
| 15 | DeepSeek V4.1 Flash (1017) (Max) | 49 | #27 (DeepSeek V4.1 Flash (1017)) |
Interactive version: theaggregate.ai/benchmark?slug=epoch-ai-mystery-game-puzzles · How It Works · Data refreshed daily, snapshot 2026-11-05.