Automated page. Written and translated automatically by AI and checked against the sources by software; no human editor. SI-IQ is an index built from public benchmark results, not an IQ test. How it works
Mystery Game Puzzles: AI model results
Mystery Game Puzzles is a Reasoning benchmark run by Epoch AI. 78 models in our data have a published result.
It is part of the SI-IQ index (loading k = 1.15, chance level 9%).
- Evaluator
- Epoch AI
- Benchmark home page
- epoch.ai/benchmarks/mystery-game-puzzles
Results
| 1 | GPT-6 Astra OpenAI | 84.0% | 150146–153 | 8 |
| 2 | GPT-6.1 Sol OpenAI | 80.0% | 148144–152 | 8 |
| 3 | Claude Opus 5.5 Anthropic | 71.0% | 144141–148 | 8 |
| 4 | Claude Sonnet 5.5 Anthropic | 65.0% | 140137–143 | 8 |
| 5 | Gemini 4 Argon Google DeepMind | 64.0% | 140137–143 | 8 |
| 6 | Claude Opus 5 Anthropic | 59.0% | 136133–140 | 8 |
| 7 | Claude Fable 5.1 Anthropic | 58.0% | 140136–143 | 8 |
| 8 | GPT-5.6 Sol OpenAI | 58.0% | 140137–143 | 8 |
| 9 | GPT-5.5 OpenAI | 56.0% | 138135–141 | 9 |
| 10 | GPT-6 Sol OpenAI | 56.0% | 140137–144 | 8 |
| 11 | Claude Fable 5 Anthropic | 52.0% | 138134–141 | 8 |
| 12 | Muse Wonder Meta AI | 51.0% | 136133–139 | 4 too little data |
| 13 | Gemini 3.8 Flash Google DeepMind | 47.0% | 130127–133 | 8 |
| 14 | DeepSeek V4 Pro 0813open DeepSeek | 43.0% | 128125–131 | 8 |
| 15 | DeepSeek V4.1 Flashopen DeepSeek | 43.0% | 128125–131 | 8 |
| 16 | Qwen 3.8 Max Alibaba | 38.0% | 130127–133 | 7 too little data |
| 17 | GPT-5.4 OpenAI | 37.0% | 131128–133 | 9 |
| 18 | Gemini 3.7 Flash Google DeepMind | 37.0% | 131128–134 | 8 |
| 19 | Claude Opus 4.8 Anthropic | 36.0% | 132129–135 | 8 |
| 20 | Claude Sonnet 5 Anthropic | 35.0% | 126123–129 | 8 |
| 21 | GPT-5.6 Terra OpenAI | 35.0% | 134131–137 | 8 |
| 22 | DeepSeek V4 Flash 0731open DeepSeek | 34.0% | 124121–127 | 8 |
| 23 | Gemini 3.1 Pro Google DeepMind | 34.0% | 127124–130 | 9 |
| 24 | Grok 4.6 xAI | 34.0% | 126123–129 | 8 |
| 25 | GLM-5.3open Z.ai (Zhipu AI) | 33.0% | 125122–128 | 8 |
| 26 | Gemini 3.5 Flash Google DeepMind | 32.0% | 126123–129 | 9 |
| 27 | Qwen3.7-Max Alibaba | 32.0% | 125122–128 | 9 |
| 28 | Claude Haiku 5.5 Anthropic | 30.0% | 129126–132 | 8 |
| 29 | Gemini 3.6 Flash Google DeepMind | 30.0% | 125122–128 | 8 |
| 30 | Grok 4.7 xAI | 29.0% | 123120–127 | 7 too little data |
| 31 | o3 OpenAI | 29.0% | 116113–119 | 10 |
| 32 | Claude Opus 4.7 Anthropic | 28.0% | 127124–130 | 9 |
| 33 | Gemini 3 Flash Google DeepMind | 26.0% | 122119–125 | 9 |
| 34 | Kimi K3open Moonshot | 26.0% | 127124–130 | 8 |
| 35 | Claude Opus 4.6 Anthropic | 25.0% | 123120–126 | 9 |
| 36 | Muse Spark 1.3 Meta AI | 25.0% | 128125–131 | 5 too little data |
| 37 | GPT-5 OpenAI | 23.0% | 121118–124 | 11 |
| 38 | GPT-5.2 OpenAI | 23.0% | 125122–128 | 10 |
| 39 | Claude Opus 4.5 Anthropic | 22.0% | 115111–118 | 10 |
| 40 | Qwen 3.6 35B-A3Bopen Alibaba | 22.0% | 111107–115 | 5 |
| 41 | Claude Opus 4.1 Anthropic | 21.0% | 105102–109 | 6 too little data |
| 42 | GPT-5.6 Luna OpenAI | 21.0% | 130127–133 | 8 |
| 43 | Nemotron 3 Ultraopen Nvidia | 20.0% | 114108–120 | 4 too little data |
| 44 | Qwen 3.5 Flash (hosted 35B-A3B) Alibaba | 20.0% | 107103–111 | 6 too little data |
| 45 | GLM-5.2open Z.ai (Zhipu AI) | 19.0% | 121118–124 | 9 |
| 46 | GPT-5.1 OpenAI | 19.0% | 120115–125 | 7 |
| 47 | Gemini 3.5 Flash-Lite Google DeepMind | 19.0% | 109106–113 | 7 |
| 48 | Qwen 3.6 Max (Preview) Alibaba | 19.0% | 118112–124 | 5 too little data |
| 49 | Kimi K2.6open Moonshot | 18.0% | 121118–124 | 9 |
| 50 | Qwen 3.6 Flash Alibaba | 18.0% | 107103–111 | 6 too little data |
| 51 | Qwen3.5 397B-A17Bopen Alibaba | 18.0% | 114109–118 | 5 |
| 52 | Claude Sonnet 4.5 Anthropic | 17.0% | 110107–113 | 11 |
| 53 | DeepSeek-V4-Proopen DeepSeek | 17.0% | 117113–120 | 9 |
| 54 | Mistral Large 4 Mistral AI | 17.0% | 122119–125 | 7 too little data |
| 55 | Qwen 3.5 Plus (hosted 397B-A17B) Alibaba | 17.0% | 109103–115 | 5 too little data |
| 56 | Qwen3.5-122B-A10Bopen Alibaba | 17.0% | 112103–121 | 3 too little data |
| 57 | Qwen3.7-Plus Alibaba | 17.0% | 115110–119 | 5 |
| 58 | Claude Sonnet 4.6 Anthropic | 16.0% | 113108–118 | 7 |
| 59 | Qwen3.7 Flash Alibaba | 15.0% | 107103–112 | 4 too little data |
| 60 | GPT-4 (Jun 2023) OpenAI | 12.0% | 5751–63 | 4 too little data |
| 61 | GPT-4o mini OpenAI | 12.0% | 6864–73 | 9 |
| 62 | Qwen 3.6 Plus Alibaba | 12.0% | 113109–117 | 7 too little data |
| 63 | GPT-5.4 Mini OpenAI | 11.0% | 117113–120 | 8 |
| 64 | GPT-5 mini OpenAI | 10.0% | 115112–118 | 11 |
| 65 | GPT-5 nano OpenAI | 9.0% | 105102–109 | 10 |
| 66 | GPT-5.4 Nano OpenAI | 9.0% | 118115–121 | 8 |
| 67 | Qwen3-235B-A22B-Thinking (Jul 2025)open Alibaba | 9.0% | 111105–118 | 5 |
| 68 | GLM-5.3-Flashopen Z.ai (Zhipu AI) | 8.0% | 120117–124 | 7 |
| 69 | MiniMax-M3open MiniMax | 8.0% | 108102–114 | 5 |
| 70 | GPT-4.1 mini OpenAI | 7.0% | 9188–95 | 10 |
| 71 | GPT-6 Luna OpenAI | 7.0% | 130127–133 | 8 |
| 72 | Qwen3.6 27Bopen Alibaba | 7.0% | 115110–119 | 5 |
| 73 | o3-mini OpenAI | 7.0% | 105101–108 | 9 |
| 74 | Inkling-Smallopen Thinking Machines | 6.0% | 118115–122 | 7 |
| 75 | Qwen3-Max Alibaba | 5.0% | 106103–110 | 6 too little data |
| 76 | o4-mini OpenAI | 5.0% | 112109–115 | 10 |
| 77 | GPT-3.5 Turbo (Jan 2024) OpenAI | 3.0% | 4742–53 | 5 too little data |
| 78 | gpt-oss-120bopen OpenAI | 2.0% | 10699–112 | 5 |
Data sources: Aider polyglot leaderboard data (Apache-2.0); Arena leaderboard dataset on Hugging Face (CC BY 4.0); Berkeley Function Calling Leaderboard, Gorilla project (Apache-2.0); Epoch AI, ‘Capabilities & benchmarking’ (CC BY 4.0); API list prices via LiteLLM's price table (MIT); each price links to the developer's pricing page; Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter..