Reasoning
Mystery Game Puzzles Leaderboard
Mystery Game Puzzles is an Epoch AI evaluation of multi-step deductive reasoning: the model works through a mystery or logic-game scenario to arrive at the correct solution. It probes sustained, structured inference rather than single-step question answering.
Source: epoch20 open models ranked+116 proprietaryData through Oct 2026
Open models ranked on Mystery Game Puzzles
# shows rank among open models / rank overall (including proprietary).
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 / 15 | DeepSeek V4.1 Flash 1017 · 763.2B | 49.0% | ~458 GB |
| 2 / 17 | DeepSeek V4 Pro 0813 · 1650.5B | 43.0% | ~991 GB |
| 3 / 18 | DeepSeek V4.1 Flash · 763.2B | 43.0% | ~458 GB |
| 4 / 27 | DeepSeek V4 Flash 0731 · 304.2B | 34.0% | ~183 GB |
| 5 / 30 | GLM 5.3 · 753.3B | 33.0% | ~456 GB |
| 6 / 47 | Kimi K3 · 2779.9B | 26.0% | ~1674 GB |
| 7 / 59 | Qwen3.6 35B A3B · 36.0B | 22.0% | ~22 GB |
| 8 / 64 | Mistral Large 4 1T A52B · 1000B | 20.0% | ~604 GB |
| 9 / 65 | NVIDIA Nemotron 3 Ultra 550B A55B BF16 · 560.5B | 20.0% | ~370 GB |
| 10 / 68 | GLM 5.2 · 753.3B | 19.0% | ~456 GB |
| 11 / 73 | Kimi K2.6 · 1026.9B | 18.0% | ~620 GB |
| 12 / 74 | Qwen3.5 397B A17B · 403.4B | 18.0% | ~242 GB |
| 13 / 77 | DeepSeek V4 Pro · 1598.8B | 17.0% | ~960 GB |
| 14 / 80 | Qwen3.5 122B A10B · 125.1B | 17.0% | ~75 GB |
| 15 / 115 | Qwen3 235B A22B Thinking 2507 · 235.1B | 9.0% | ~142 GB |
| 16 / 116 | GLM 5.3 Flash · 321.3B | 8.0% | ~195 GB |
| 17 / 119 | MiniMax M3 · 427.0B | 8.0% | ~257 GB |
| 18 / 125 | Qwen3.6 27B · 27.8B | 7.0% | ~17 GB |
| 19 / 127 | Inkling Small · 266.0B | 6.0% | ~160 GB |
| 20 / 136 | GPT OSS 120B · 116.8B | 2.0% | ~71 GB |
Score vs model size
Which models give the most quality for their size — the ones worth running locally.
- Qwen3.6 27B, 28B, score 7.0% — on the efficiency frontier (best score at its size or smaller).
- Qwen3.6 35B A3B, 36B, score 22.0% — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4 Flash 0731, 304B, score 34.0% — on the efficiency frontier (best score at its size or smaller).
- DeepSeek V4.1 Flash 1017, 763B, score 49.0% — on the efficiency frontier (best score at its size or smaller).
Mystery Game Puzzles: frequently asked questions
- What is the best open LLM on Mystery Game Puzzles?
- DeepSeek V4.1 Flash 1017 is the top open model on Mystery Game Puzzles, scoring 49.0%. Among all models tested — including proprietary ones — it ranks #15. The top model overall is GPT 6 Astra Max (OpenAI) at 84.0%.
- What's the best Mystery Game Puzzles model you can run on a 24 GB GPU?
- Qwen3.6 35B A3B is the highest-scoring open model that fits in 24 GB at 4-bit quantization (~22 GB), scoring 22.0% on Mystery Game Puzzles.
- Can open models match proprietary models on Mystery Game Puzzles?
- Not quite on Mystery Game Puzzles: the strongest proprietary model (GPT 6 Astra Max) scores 84.0%, ahead of the best open model (DeepSeek V4.1 Flash 1017) at 49.0% — but you can run the open one yourself.
Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.