Inference
Epoch AI

Mystery Game Puzzles

Evaluated by Epoch AI on their own harness.

ModelsModels with a result.62
Top resultBest result on this test.84.0%GPT-6 Astra
Top-3 spreadPoints from first to third.26.0 pts
YearYear of release.N/A

Result and price

020406080100$0.1$0.3$1$3$10$30$100RESULTOUTPUT PRICE PER 1M TOKENS · LOG SCALE

Results

62 results
#ModelResultThe score from the source.
01GPT-6 Astramax effort84.0% ±3.7
02Claude Opus 5max effort59.0% ±4.9
03Claude Fable 5.1max effort58.0% ±5.0
03GPT-5.6 Solmax effort58.0% ±5.0
05GPT-5.5xhigh effort56.0% ±5.0
06Claude Fable 5max effort52.0% ±5.0
07Gemini 3.8 Flashhigh effort47.0% ±5.0
08Qwen3.8 Maxxhigh effort38.0% ±4.9
09Gemini 3.7 Flashhigh effort37.0% ±4.9
09GPT-5.4xhigh effort37.0% ±4.9
11Claude Opus 4.8max effort36.0% ±4.8
12GPT-5.6 Terramax effort35.0% ±4.8
12Claude Sonnet 5max effort35.0% ±4.8
14Grok 4.6xhigh effort34.0% ±4.8
14Gemini 3.1 Prohigh effort34.0% ±4.8
16GLM-5.3max effort33.0% ±4.7
17Gemini 3.5 Flashhigh effort32.0% ±4.7
18Gemini 3.6 Flashhigh effort30.0% ±4.6
19o3high effort29.0% ±4.6
20Claude Opus 4.7max effort28.0% ±4.5
21Qwen3.7 Maxnone effort26.0% ±4.4
22Muse Spark 1.3max effort25.0% ±4.5
22Claude Opus 4.6max effort25.0% ±4.4
24GPT-5.2high effort23.0% ±4.2
24GPT-5high effort23.0% ±4.2
26Claude Opus 4.522.0% ±4.2
26Qwen3.6-35B-A3Bnone effort22.0% ±4.2
28GPT-5.6 Lunamax effort21.0% ±4.1
28Claude Opus 4.121.0% ±4.1
30Gemini 3 Flashhigh effort20.0% ±4.0
31Qwen 3.6 Max (preview)none effort19.0% ±3.9
31Gemini 3.5 Flash-Litelow effort19.0% ±3.9
33Qwen3.5 397B A17Bnone effort18.0% ±3.9
33Qwen3.6 Flashnone effort18.0% ±3.9
35DeepSeek V4 Pro 0423none effort17.0% ±3.8
35Qwen3.7 Plus17.0% ±3.8
35Qwen3.5-122B-A10Bnone effort17.0% ±3.8
35Claude Sonnet 4.517.0% ±3.8
35Qwen 3.5 Plus (hosted 397B-A17B)none effort17.0% ±3.8
40Claude Sonnet 4.6low effort16.0% ±3.7
40GPT-5.1medium effort16.0% ±3.7
40Qwen 3.5 Flash (hosted 35B-A3B)none effort16.0% ±3.7
43GLM-5.2medium effort15.0% ±3.6
44Qwen3.7 Flashnone effort14.0% ±3.5
45Qwen3.6 Plusnone effort12.0% ±3.3
45GPT-4o mini12.0% ±3.3
45GPT-4 (Jun 2023)12.0% ±3.3
48Qwen3-235B-A22B (Jul 2025)9.0% ±2.9
49GLM-5.3-Flashmax effort8.0% ±2.7
49MiniMax M3none effort8.0% ±2.7
49GPT-5 nanohigh effort8.0% ±2.7
52GPT-5.4 minimedium effort7.0% ±2.6
52Qwen3.6-27Bnone effort7.0% ±2.6
52o3-minihigh effort7.0% ±2.6
52GPT-4.1 mini7.0% ±2.6
56Inkling-Smallxhigh effort6.0% ±2.4
57GPT-5.4 nanohigh effort5.0% ±2.2
57GPT-5 minihigh effort5.0% ±2.2
57o4-mini (high)high effort5.0% ±2.2
57Qwen3 Max5.0% ±2.2
61GPT-3.5 Turbo3.0% ±1.7
62GPT-OSS 120Bhigh effort0.0%