# Mystery Game Puzzles (Epoch AI) (Mystery Game Puzzles)

> Epoch AI's Mystery Game Puzzles benchmark asks models for the single best next move in 100 positions from a puzzle variant of a well-known game whose identity Epoch keeps secret.

Canonical page: https://benchlm.ai/benchmarks/mysterygamepuzzles

- Category: [Reasoning](/reasoning)
- Last updated: November 13, 2026

## About Mystery Game Puzzles

- Year: 2026
- Tasks: 100 single-best-move positions from an undisclosed game
- Format: Exact-match move answer from text input, minimal agent scaffold
- Difficulty: Unfamiliar game positions, single best move
- Paper: [Mystery Game Puzzles (Epoch AI)](https://epoch.ai/benchmarks/mystery-game-puzzles)

The setup mirrors Epoch's Chess Puzzles: positions come from random games rolled forward to mid-game and filtered to states with one best move. Models receive the state as text in a minimal agent scaffold and submit a move through a tool. Answers are normalized to the game's standard notation and must match the key; an illegal move is wrong and an answer with no valid move gets no credit. Epoch does not publish the prompt, examples or transcripts, to keep the game hidden and make benchmark-specific preparation harder.

Mystery Game Puzzles is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (76 models)

| Rank | Model | Creator | Score | Who ran it |
|------|-------|---------|-------|-----------|
| 1 | [Claude Fable 5.2](/models/claude-fable-5-2) | Anthropic | 90.0% | Run by Mystery Game Puzzles |
| 2 | [GPT-6 Astra](/models/gpt-6-astra) | OpenAI | 84.0% | Run by Mystery Game Puzzles |
| 3 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | OpenAI | 80.0% | Run by Mystery Game Puzzles |
| 4 | [GPT-6.1 Terra](/models/gpt-6-1-terra) | OpenAI | 74.0% | Run by Mystery Game Puzzles |
| 5 | [Gemini 4 Carbon](/models/gemini-4-carbon) | Google | 73.0% | Run by Mystery Game Puzzles |
| 6 | [Claude Opus 5.5](/models/claude-opus-5-5) | Anthropic | 71.0% | Run by Mystery Game Puzzles |
| 7 | [Claude Sonnet 5.5](/models/claude-sonnet-5-5) | Anthropic | 65.0% | Run by Mystery Game Puzzles |
| 8 | [Gemini 4 Argon](/models/gemini-4-argon) | Google | 64.0% | Run by Mystery Game Puzzles |
| 9 | [Claude Opus 5](/models/claude-opus-5) | Anthropic | 59.0% | Run by Mystery Game Puzzles |
| 10 | [Claude Fable 5.1](/models/claude-fable-5-1) | Anthropic | 58.0% | Run by Mystery Game Puzzles |
| 11 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | OpenAI | 58.0% | Run by Mystery Game Puzzles |
| 12 | [GPT-6 Sol](/models/gpt-6-sol) | OpenAI | 56.0% | Run by Mystery Game Puzzles |
| 13 | [GPT-5.5](/models/gpt-5-5) | OpenAI | 56.0% | Run by Mystery Game Puzzles |
| 14 | [Claude Fable 5](/models/claude-fable) | Anthropic | 52.0% | Run by Mystery Game Puzzles |
| 15 | [Muse Wonder](/models/muse-wonder) | Meta | 51.0% | Run by Mystery Game Puzzles |
| 16 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | DeepSeek | 49.0% | Run by Mystery Game Puzzles |
| 17 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | Google | 47.0% | Run by Mystery Game Puzzles |
| 18 | [DeepSeek V4 Pro 0813](/models/deepseek-v4-pro-0813) | DeepSeek | 43.0% | Run by Mystery Game Puzzles |
| 19 | [Kimi K3.1 Pro](/models/kimi-3-1-pro) | Moonshot AI | 42.0% | Run by Mystery Game Puzzles |
| 20 | [Qwen3.8 Max](/models/qwen3-8-max) | Alibaba | 38.0% | Run by Mystery Game Puzzles |
| 21 | [Gemini 3.7 Flash](/models/gemini-3-7-flash) | Google | 37.0% | Run by Mystery Game Puzzles |
| 22 | [GPT-5.4](/models/gpt-5-4) | OpenAI | 37.0% | Run by Mystery Game Puzzles |
| 23 | [Claude Opus 4.8](/models/claude-opus-4-8) | Anthropic | 36.0% | Run by Mystery Game Puzzles |
| 24 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | OpenAI | 35.0% | Run by Mystery Game Puzzles |
| 25 | [Claude Sonnet 5](/models/claude-sonnet-5) | Anthropic | 35.0% | Run by Mystery Game Puzzles |
| 26 | [Grok 4.6](/models/grok-4-6) | xAI | 34.0% | Run by Mystery Game Puzzles |
| 27 | [Gemini 3.1 Pro](/models/gemini-3-1-pro) | Google | 34.0% | Run by Mystery Game Puzzles |
| 28 | [DeepSeek V4 Flash 0731](/models/deepseek-v4-flash-0731) | DeepSeek | 34.0% | Run by Mystery Game Puzzles |
| 29 | [GLM-5.3](/models/glm-5-3) | Z.AI | 33.0% | Run by Mystery Game Puzzles |
| 30 | [Gemini 3.5 Flash](/models/gemini-3-5-flash) | Google | 32.0% | Run by Mystery Game Puzzles |
| 31 | [Qwen3.7 Max](/models/qwen3-7-max) | Alibaba | 32.0% | Run by Mystery Game Puzzles |
| 32 | [Claude Haiku 5.5](/models/claude-haiku-5-5) | Anthropic | 30.0% | Run by Mystery Game Puzzles |
| 33 | [Gemini 3.6 Flash](/models/gemini-3-6-flash) | Google | 30.0% | Run by Mystery Game Puzzles |
| 34 | [o3](/models/o3) | OpenAI | 29.0% | Run by Mystery Game Puzzles |
| 35 | [Grok 4.7](/models/grok-4-7) | xAI | 29.0% | Run by Mystery Game Puzzles |
| 36 | [Claude Opus 4.7](/models/claude-opus-4-7) | Anthropic | 28.0% | Run by Mystery Game Puzzles |
| 37 | [Kimi K3](/models/kimi-k3) | Moonshot AI | 26.0% | Run by Mystery Game Puzzles |
| 38 | [Claude Opus 4.6](/models/claude-opus-4-6) | Anthropic | 25.0% | Run by Mystery Game Puzzles |
| 39 | [Muse Spark 1.3](/models/muse-spark-1-3) | Meta | 25.0% | Run by Mystery Game Puzzles |
| 40 | [GPT-5.2](/models/gpt-5-2) | OpenAI | 23.0% | Run by Mystery Game Puzzles |
| 41 | [GPT-5 (high)](/models/gpt-5-high) | OpenAI | 23.0% | Run by Mystery Game Puzzles |
| 42 | [Claude Opus 4.5](/models/claude-opus-4-5) | Anthropic | 22.0% | Run by Mystery Game Puzzles |
| 43 | [Qwen3.6-35B-A3B](/models/qwen3-6-35b-a3b) | Alibaba | 22.0% | Run by Mystery Game Puzzles |
| 44 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | OpenAI | 21.0% | Run by Mystery Game Puzzles |
| 45 | [Gemini 3 Flash](/models/gemini-3-flash) | Google | 20.0% | Run by Mystery Game Puzzles |
| 46 | [Nemotron 3 Ultra](/models/nemotron-3-ultra) | NVIDIA | 20.0% | Run by Mystery Game Puzzles |
| 47 | [Qwen3.5 Flash](/models/qwen3-5-flash) | Alibaba | 20.0% | Run by Mystery Game Puzzles |
| 48 | [Qwen 3.6 Max (preview)](/models/qwen3-6-max-preview) | Alibaba | 19.0% | Run by Mystery Game Puzzles |
| 49 | [Gemini 3.5 Flash-Lite](/models/gemini-3-5-flash-lite) | Google | 19.0% | Run by Mystery Game Puzzles |
| 50 | [Kimi K2.6](/models/kimi-2-6) | Moonshot AI | 18.0% | Run by Mystery Game Puzzles |
| 51 | [Qwen3.5 397B](/models/qwen3-5-397b) | Alibaba | 18.0% | Run by Mystery Game Puzzles |
| 52 | [Qwen3.7 Plus](/models/qwen3-7-plus) | Alibaba | 17.0% | Run by Mystery Game Puzzles |
| 53 | [Mistral Large 4](/models/mistral-large-4) | Mistral | 17.0% | Run by Mystery Game Puzzles |
| 54 | [Qwen3.5-122B-A10B](/models/qwen3-5-122b-a10b) | Alibaba | 17.0% | Run by Mystery Game Puzzles |
| 55 | [Qwen3.5 Plus](/models/qwen3-5-plus) | Alibaba | 17.0% | Run by Mystery Game Puzzles |
| 56 | [Claude Sonnet 4.5](/models/claude-sonnet-4-5) | Anthropic | 17.0% | Run by Mystery Game Puzzles |
| 57 | [GPT-5.1](/models/gpt-5-1) | OpenAI | 16.0% | Run by Mystery Game Puzzles |
| 58 | [Claude Sonnet 4.6](/models/claude-sonnet-4-6) | Anthropic | 16.0% | Run by Mystery Game Puzzles |
| 59 | [GLM-5.2](/models/glm-5-2) | Z.AI | 15.0% | Run by Mystery Game Puzzles |
| 60 | [GPT-5 (medium)](/models/gpt-5-medium) | OpenAI | 15.0% | Run by Mystery Game Puzzles |
| 61 | [Qwen3.7 Flash](/models/qwen3-7-flash) | Alibaba | 15.0% | Run by Mystery Game Puzzles |
| 62 | [Qwen3.6 Plus](/models/qwen3-6-plus) | Alibaba | 12.0% | Run by Mystery Game Puzzles |
| 63 | [GPT-4o mini](/models/gpt-4o-mini) | OpenAI | 12.0% | Run by Mystery Game Puzzles |
| 64 | [Qwen3 235B 2507 (Reasoning)](/models/qwen3-235b-2507-reasoning) | Alibaba | 9.0% | Run by Mystery Game Puzzles |
| 65 | [GLM-5.3-Flash](/models/glm-5-3-flash) | Z.AI | 8.0% | Run by Mystery Game Puzzles |
| 66 | [MiniMax M3](/models/minimax-m3) | MiniMax | 8.0% | Run by Mystery Game Puzzles |
| 67 | [GPT-6 Luna](/models/gpt-6-luna) | OpenAI | 7.0% | Run by Mystery Game Puzzles |
| 68 | [GPT-5.4 mini](/models/gpt-5-4-mini) | OpenAI | 7.0% | Run by Mystery Game Puzzles |
| 69 | [Qwen3.6-27B](/models/qwen3-6-27b) | Alibaba | 7.0% | Run by Mystery Game Puzzles |
| 70 | [o3-mini](/models/o3-mini) | OpenAI | 7.0% | Run by Mystery Game Puzzles |
| 71 | [GPT-4.1 mini](/models/gpt-4-1-mini) | OpenAI | 7.0% | Run by Mystery Game Puzzles |
| 72 | [Inkling-Small](/models/inkling-small) | Thinking Machines Lab | 6.0% | Run by Mystery Game Puzzles |
| 73 | [GPT-5.4 nano](/models/gpt-5-4-nano) | OpenAI | 5.0% | Run by Mystery Game Puzzles |
| 74 | [Qwen3 Max](/models/qwen3-max) | Alibaba | 5.0% | Run by Mystery Game Puzzles |
| 75 | [o4-mini (high)](/models/o4-mini-high) | OpenAI | 5.0% | Run by Mystery Game Puzzles |
| 76 | [GPT-OSS 120B](/models/gpt-oss-120b) | OpenAI | 0.0% | Run by Mystery Game Puzzles |

## FAQ

### Why is the game secret?

Epoch keeps the game's identity undisclosed so labs cannot prepare for the benchmark specifically. It is otherwise built like Epoch's Chess Puzzles.

### Which model leads Mystery Game Puzzles?

Claude Fable 5.2 at max effort leads with 90%, ahead of GPT-6 Astra (84%), GPT-6.1 Sol (80%) and GPT-6.1 Terra (74%).

## Compare Top Models on Mystery Game Puzzles

- [Claude Fable 5.2 vs GPT-6 Astra](/compare/claude-fable-5-2-vs-gpt-6-astra)
- [GPT-6 Astra vs GPT-6.1 Sol](/compare/gpt-6-1-sol-vs-gpt-6-astra)
- [GPT-6.1 Sol vs GPT-6.1 Terra](/compare/gpt-6-1-sol-vs-gpt-6-1-terra)
- [GPT-6.1 Terra vs Gemini 4 Carbon](/compare/gemini-4-carbon-vs-gpt-6-1-terra)