Benchmarks · reasoning
Mystery Game Puzzles (Epoch AI)
Epoch AI's deduction puzzles: work out the hidden rules of a mystery game from its observations. Score is the share solved. Data: Epoch AI, CC BY.
79 models · 136 results · data as of 2026-10-29 · epoch.ai ↗
Leaderboard
GPT-6 Astra max
All results
136 rows| # | Model | Effort | Score | Harness |
|---|---|---|---|---|
| 1 | GPT-6 Astra | max | 84.0% | Epoch AI |
| 2 | GPT-6.1 Sol | max | 80.0% | Epoch AI |
| 3 | GPT-6.1 Terra | max | 74.0% | Epoch AI |
| 4 | Opus 5.5 | max | 71.0% | Epoch AI |
| 5 | Sonnet 5.5 | max | 65.0% | Epoch AI |
| 6 | Gemini 4 Argon | high | 64.0% | Epoch AI |
| 7 | Fable 5.1 | max | 58.0% | Epoch AI |
| 8 | GPT-5.6 Sol | max | 58.0% | Epoch AI |
| 9 | GPT-5.5 | xhigh | 56.0% | Epoch AI |
| 10 | GPT-6 Sol | max | 56.0% | Epoch AI |
| 11 | Fable 5 | max | 52.0% | Epoch AI |
| 12 | GPT-5.5 | high | 52.0% | Epoch AI |
| 13 | DeepSeek V4.1 Flash 1017 | max | 49.0% | Epoch AI |
| 14 | Gemini 3.8 Flash | high | 47.0% | Epoch AI |
| 15 | DeepSeek V4 Pro 0813 | max | 43.0% | Epoch AI |
| 16 | DeepSeek V4.1 Flash | max | 43.0% | Epoch AI |
| 17 | Qwen 3.8 Max | xhigh | 38.0% | Epoch AI |
| 18 | Opus 5 | max | 37.0% | Epoch AI |
| 19 | Gemini 3.7 Flash | high | 37.0% | Epoch AI |
| 20 | GPT-5.4 | xhigh | 37.0% | Epoch AI |
| 21 | Opus 4.8 | max | 36.0% | Epoch AI |
| 22 | GPT-5.6 Terra | max | 35.0% | Epoch AI |
| 23 | DeepSeek V4 Flash 0731 | max | 34.0% | Epoch AI |
| 24 | Gemini 3.1 Pro | high | 34.0% | Epoch AI |
| 25 | Grok 4.6 | xhigh | 34.0% | Epoch AI |
| 26 | GLM 5.3 | max | 33.0% | Epoch AI |
| 27 | GPT-5.6 Sol | none | 33.0% | Epoch AI |
| 28 | Gemini 3.1 Pro | medium | 32.0% | Epoch AI |
| 29 | Gemini 3.5 Flash | high | 32.0% | Epoch AI |
| 30 | Qwen 3.7 Max | max | 32.0% | Epoch AI |