Inference
EN
Epoch AI

Mystery Game Puzzles

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.80
Top resultBest result on this benchmark.90.0%Claude Fable 5.2
Top-3 spreadPoints from first to third.10.0 pts
YearYear of release.—

Result and price

75 of 80 results

Results

80 results
#ModelResultThe score from the source.
01Claude Fable 5.2max effort90.0% ±3.0
02GPT-6 Astramax effortAbove fitFar above this model's index.84.0% ±3.7
03GPT-6.1 Solmax effortAbove fitFar above this model's index.80.0% ±4.0
04GPT-6.1 Terramax effortAbove fitFar above this model's index.74.0% ±4.4
05Gemini 4 Carbonhigh effort73.0% ±4.5
06Claude Opus 5.5max effort71.0% ±4.6
07Claude Sonnet 5.5max effort65.0% ±4.8
08Gemini 4 Argonhigh effort64.0% ±4.8
09Claude Opus 5max effort59.0% ±4.9
10Claude Fable 5.1max effort58.0% ±5.0
10GPT-5.6 Solmax effort58.0% ±5.0
12GPT-6 Solmax effort56.0% ±5.0
12GPT-5.5xhigh effort56.0% ±5.0
14Claude Fable 5max effort52.0% ±5.0
15Muse Wondermax effort51.0% ±5.0
16Gemini 3.8 Flashhigh effort47.0% ±5.0
17DeepSeek V4.1 Flash 1017max effort43.0% ±5.0
17DeepSeek V4 Pro 0813max effort43.0% ±5.0
19Qwen3.8 Maxxhigh effort38.0% ±4.9
20Gemini 3.7 Flashhigh effort37.0% ±4.9
20GPT-5.4xhigh effort37.0% ±4.9
22Claude Opus 4.8max effort36.0% ±4.8
23GPT-5.6 Terramax effort35.0% ±4.8
23Claude Sonnet 5max effort35.0% ±4.8
25Grok 4.6xhigh effort34.0% ±4.8
25Gemini 3.1 Prohigh effort34.0% ±4.8
25DeepSeek V4 Flash 0731max effort34.0% ±4.8
28GLM-5.3max effort33.0% ±4.7
29Gemini 3.5 Flashhigh effort32.0% ±4.7
30Claude Haiku 5.5max effort30.0% ±4.6
30Gemini 3.6 Flashhigh effort30.0% ±4.6
32Grok 4.7xhigh effort29.0% ±4.6
32o3high effort29.0% ±4.6
34Claude Opus 4.7max effort28.0% ±4.5
35Kimi K3max effort26.0% ±4.4
35Qwen3.7 Maxnone effort26.0% ±4.4
37Muse Spark 1.3max effort25.0% ±4.5
37Claude Opus 4.6max effort25.0% ±4.4
39GPT-5.2high effort23.0% ±4.2
39GPT-5high effort23.0% ±4.2
41Claude Opus 4.522.0% ±4.2
41Qwen3.6-35B-A3Bnone effort22.0% ±4.2
43GPT-5.6 Lunamax effort21.0% ±4.1
43Claude Opus 4.121.0% ±4.1
45Gemini 3 Flashhigh effort20.0% ±4.0
46Qwen 3.6 Max (preview)none effort19.0% ±3.9
46Gemini 3.5 Flash-Litelow effort19.0% ±3.9
48Mistral Large 4high effort18.5% ±4.0
49Qwen3.5 397B A17Bnone effort18.0% ±3.9
49Qwen3.6 Flashnone effort18.0% ±3.9
51Qwen3.7 Plusnone effort17.0% ±3.8
51DeepSeek V4 Pro 0423none effort17.0% ±3.8
51Qwen3.5-122B-A10Bnone effort17.0% ±3.8
51Claude Sonnet 4.517.0% ±3.8
51Qwen 3.5 Plus (hosted 397B-A17B)none effort17.0% ±3.8
56Claude Sonnet 4.6low effort16.0% ±3.7
56GPT-5.1medium effort16.0% ±3.7
56Qwen 3.5 Flash (hosted 35B-A3B)none effort16.0% ±3.7
59GLM-5.2medium effort15.0% ±3.6
60Qwen3.7 Flashnone effort14.0% ±3.5
61Kimi K2.6none effort12.0% ±3.3
61Qwen3.6 Plusnone effort12.0% ±3.3
61GPT-4o mini12.0% ±3.3
61GPT-4 (Jun 2023)12.0% ±3.3
65Qwen3-235B-A22B (Jul 2025)9.0% ±2.9
66GLM-5.3-Flashmax effort8.0% ±2.7
66MiniMax M3none effort8.0% ±2.7
66GPT-5 nanohigh effort8.0% ±2.7
69GPT-6 Lunamax effort7.0% ±2.6
69GPT-5.4 minimedium effort7.0% ±2.6
69Qwen3.6-27Bnone effort7.0% ±2.6
69o3-minihigh effort7.0% ±2.6
69GPT-4.1 mini7.0% ±2.6
74Inkling-Smallxhigh effort6.0% ±2.4
75GPT-5.4 nanohigh effort5.0% ±2.2
75GPT-5 minihigh effort5.0% ±2.2
75Qwen3 Max5.0% ±2.2
75o4-minihigh effort5.0% ±2.2
79GPT-3.5 Turbo3.0% ±1.7
80GPT-OSS 120Bhigh effort0.0%