Mystery Game Puzzles
Deduction puzzles in the style of mystery games. Run by Epoch AI.
As of 19 Sept 2026, GPT-6 Astra leads Mystery Game Puzzles on BenchLeader with 84.0%, ahead of Claude Opus 5 at 59.0%, across 128 model configurations with a published result.
- Published by
- Epoch AI Benchmarking Hub
- Category
- Reasoning
- Index weight
- Reference only
- Models
- 128
- Data as of
- 19 Sept 2026
CC BY 4.0 — Epoch AI, ‘AI Benchmarking Hub’, epoch.ai/benchmarks. Mirrored boards credit their original publishers.
What the test looks like
Deduction puzzles with clues and suspects; the model must reason to the unique solution.
How it is scored
Percent solved, run by Epoch AI.
What to keep in mind
Puzzle styles are specific; models can be tuned to them.
- 1GPT-6 Astra (max)84.0%
- 2Claude Opus 5 (max)59.0%
- 3Claude Fable 5.1 (max)58.0%
- 4GPT-5.6 Sol (max)58.0%
- 5GPT-5.5 (xhigh)56.0%
- 6GPT-5.5 (high)52.0%
- 7Claude Fable 5 (max)52.0%
- 8Gemini 3.8 Flash (high)47.0%
- 9DeepSeek V4 Pro 0813 (max)43.0%
- 10Qwen3 8 (xhigh)38.0%
- 11Claude Opus 537.0%
- 12GPT-5.4 (xhigh)37.0%
- 13Gemini 3.7 Flash (high)37.0%
- 14Claude Opus 4.8 (max)36.0%
- 15GPT-5.6 Terra (max)35.0%
128 of 128
| # | ||||
|---|---|---|---|---|
| 1 |  GPT-6 AstramaxOpenAI | 84.0% | 71.8 | 2026-09-03 |
| 2 |  Claude Opus 5maxAnthropic | 59.0% | 69.9 | 2026-07-24 |
| 3 | Claude Fable 5.1maxAnthropic | 58.0% | 69.7 | 2026-09-01 |
| 4 | GPT-5.6 SolmaxOpenAI | 58.0% | 68.8 | 2026-07-09 |
| 5 | GPT-5.5xhighOpenAI | 56.0% | 67.6 | 2026-04-23 |
| 6 | GPT-5.5highOpenAI | 52.0% | 67.0 | 2026-04-23 |
| 7 | Claude Fable 5maxAnthropic | 52.0% | 65.8 | 2026-06-09 |
| 8 |  Gemini 3.8 FlashhighGoogle | 47.0% | 64.5 | 2026-09-02 |
| 9 |  DeepSeek V4 Pro 0813maxDeepSeekopen ↗ | 43.0% | 56.0 | 2026-08-13 |
| 10 |  Qwen3 8xhighAlibaba | 38.0% | 57.7 | 2026-08-02 |
| 11 | Claude Opus 5Anthropic | 37.0% | 66.7 | 2026-07-24 |
| 12 | GPT-5.4xhighOpenAI | 37.0% | 65.4 | 2026-03-05 |
| 13 | Gemini 3.7 FlashhighGoogle | 37.0% | 64.6 | 2026-08-13 |
| 14 | Claude Opus 4.8maxAnthropic | 36.0% | 64.2 | 2026-05-28 |
| 15 | GPT-5.6 TerramaxOpenAI | 35.0% | 65.0 | 2026-07-09 |
| 16 | Claude Sonnet 5maxAnthropic | 35.0% | 60.5 | 2026-06-30 |
| 17 |  Grok 4.6xhighxAI | 34.0% | 64.6 | 2026-08-12 |
| 18 | Gemini 3.1 ProhighGoogle | 34.0% | 60.7 | 2026-02-19 |
| 19 | DeepSeek V4 Flash 0731maxDeepSeekopen ↗ | 34.0% | 54.4 | 2026-07-31 |
| 20 |  GLM-5.3maxZhipu AIopen ↗ | 33.0% | 65.8 | 2026-08-14 |
| 21 | GPT-5.6 Solno reasoningOpenAI | 33.0% | 55.2 | 2026-07-09 |
| 22 | Gemini 3.5 FlashhighGoogle | 32.0% | 63.6 | 2026-05-19 |
| 23 | Qwen3 7maxAlibaba | 32.0% | 61.0 | 2026-05-19 |
| 24 | Gemini 3.1 PromediumGoogle | 32.0% | – | 2026-02-19 |
| 25 | Claude Opus 4.8xhighAnthropic | 31.0% | – | 2026-05-28 |
| 26 | Gemini 3.6 FlashhighGoogle | 30.0% | 61.7 | 2026-07-21 |
| 27 | o3highOpenAI | 29.0% | 53.0 | 2025-04-16 |
| 28 | Gemini 3.1 ProlowGoogle | 29.0% | – | 2026-02-19 |
| 29 | Claude Opus 4.7maxAnthropic | 28.0% | 64.3 | 2026-04-16 |
| 30 | GPT-5.5lowOpenAI | 28.0% | 61.2 | 2026-04-23 |
| 31 | GPT-5.4mediumOpenAI | 28.0% | 55.9 | 2026-03-05 |
| 32 | Gemini 3.5 FlashlowGoogle | 28.0% | – | 2026-05-19 |
| 33 |  Kimi K3maxMoonshot AIopen ↗ | 26.0% | 67.2 | 2026-07-16 |
| 34 | GPT-5.6 SollowOpenAI | 26.0% | 63.7 | 2026-07-09 |
| 35 | Gemini 3 FlashlowGoogle | 26.0% | – | 2025-12-17 |
| 36 | Qwen3 7no reasoningAlibaba | 26.0% | – | 2026-05-19 |
| 37 | Claude Opus 4.6maxAnthropic | 25.0% | 59.2 | 2026-02-05 |
| 38 | Gemini 3 FlashminimalGoogle | 25.0% | 54.9 | 2025-12-17 |
| 39 | Gemini 3.6 FlashminimalGoogle | 25.0% | – | 2026-07-21 |
| 40 | GPT-5highOpenAI | 23.0% | 58.6 | 2025-08-07 |
| 41 | GPT-5.2highOpenAI | 23.0% | 55.7 | 2025-12-11 |
| 42 | o3mediumOpenAI | 23.0% | 53.7 | 2025-04-16 |
| 43 | Gemini 3.6 FlashlowGoogle | 23.0% | – | 2026-07-21 |
| 44 | GPT-5.2mediumOpenAI | 22.0% | 58.7 | 2025-12-11 |
| 45 | Claude Opus 4.5Anthropic | 22.0% | 58.2 | 2025-11-24 |
| 46 | Qwen3.6 35B-A3Bno reasoningAlibaba | 22.0% | 43.6 | 2026-04-14 |
| 47 | GPT-5.6 LunamaxOpenAI | 21.0% | 60.2 | 2026-07-09 |
| 48 | Claude Opus 4.1Anthropic | 21.0% | 51.9 | 2025-08-05 |
| 49 | Gemini 3 FlashhighGoogle | 20.0% | 58.7 | 2025-12-17 |
| 50 | Qwen3.5 FlashAlibaba | 20.0% | 52.5 | 2026-02-25 |
| 51 |  Nemotron 3 UltraNVIDIA | 20.0% | 48.3 | 2026-06-04 |
| 52 | GPT-5.6 Lunano reasoningOpenAI | 20.0% | 46.0 | 2026-07-09 |
| 53 | GPT-5.6 TerralowOpenAI | 19.0% | 57.8 | 2026-07-09 |
| 54 | Gemini 3.5 Flash LitelowGoogle | 19.0% | – | 2026-07-21 |
| 55 | GLM-5.2lowZhipu AIopen ↗ | 19.0% | – | 2026-06-16 |
| 56 | GPT-5.1lowOpenAI | 19.0% | – | 2025-11-13 |
| 57 | Qwen3 6no reasoningAlibaba | 19.0% | – | 2026-04-20 |
| 58 | Kimi K2.6Moonshot AIopen ↗ | 18.0% | 60.5 | 2026-04-20 |
| 59 | GPT-5.5no reasoningOpenAI | 18.0% | 54.0 | 2026-04-23 |
| 60 | Qwen3.5 397B-A17Bno reasoningAlibabaopen ↗ | 18.0% | 51.7 | 2026-02-13 |
Cite as: BenchLeader, “Mystery Game Puzzles leaderboard”, https://www.benchleader.com/benchmarks/mystery_game_puzzles, data as of 19 Sept 2026.
Mystery Game Puzzles: questions
- What does Mystery Game Puzzles measure?
- Deduction puzzles with clues and suspects; the model must reason to the unique solution. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads Mystery Game Puzzles?
- GPT-6 Astra leads Mystery Game Puzzles with 84.0% as of 19 Sept 2026, ahead of Claude Opus 5 at 59.0%.
- How many models have Mystery Game Puzzles results?
- 128 model configurations have a Mystery Game Puzzles result on BenchLeader, all taken from Epoch AI Benchmarking Hub.
- Who runs Mystery Game Puzzles and how often is it updated?
- Mystery Game Puzzles is published by Epoch AI Benchmarking Hub. BenchLeader re-reads the published results every morning and records the date each result was published.
- Does Mystery Game Puzzles count toward the BenchLeader Index?
- No. Mystery Game Puzzles is shown for reference but left out of the composite index.