modelbenchmark.io
All benchmarks

Mystery Game Puzzles

Reasoning · questions rotate

Multi-step deduction inside a game the model must reason about from its rules.

How far to trust it

Newly added and rotating, so contamination is low. The spread between models is small, which limits how much it can say.

Measured

Models scored
98
Spread between models
17.1
standard deviation, points
Measurement noise
3.9
published stderr
Weight
0.95
share of spread that is signal

The weight is (spread² − noise²) / spread². It is computed, never chosen. Most of this benchmark's spread is real, so it still separates models.

Every model scored

the median across configurations, so one heroic run cannot lead

View
RankModelLabScore
1GPT-6 AstraOpenAI84.0
2GPT-6.1 SolOpenAI80.0
3Claude Opus 5.5Anthropic71.0
4Claude Sonnet 5.5Anthropic65.0
5Gemini 4 ArgonGoogle64.0
6Claude Fable 5.1Anthropic58.0
7GPT-6 SolOpenAI56.0
8Claude Fable 5Anthropic52.0
9GPT-5.5OpenAI52.0
10Muse WonderMeta51.0
11Claude Opus 5Anthropic48.0
12Gemini 3.8 FlashGoogle47.0
13DeepSeek V4.1 FlashDeepSeek43.0
14DeepSeek V4 Pro 0813DeepSeek43.0
15GPT-5.6 SolOpenAI42.0
16Qwen3.8 MaxAlibaba38.0
17Gemini 3.7 FlashGoogle37.0
18Claude Sonnet 5Anthropic35.0
19DeepSeek V4 Flash 0731DeepSeek34.0
20Grok 4.6xAI34.0
21Claude Opus 4.8Anthropic33.5
22GLM-5.3Zhipu33.0
23GPT-5.6 SolOpenAI33.0
24Gemini 3.1 Pro PreviewGoogle32.0
25Claude Haiku 5.5Anthropic30.0
26Gemini 3.5 FlashGoogle30.0
27Grok 4.7xAI29.0
28Qwen3.7-MaxAlibaba29.0
29Claude Opus 4.7Anthropic28.0
30GPT-5.4OpenAI28.0
31o3OpenAI26.0
32Gemini 3.6 FlashGoogle25.0
33Gemini 3 Flash PreviewGoogle25.0
34Claude Opus 4.5Anthropic22.0
35GPT 5.2OpenAI22.0
36Qwen3.6 35B-A3BAlibaba22.0
37Claude Opus 4.1Anthropic21.0
38GPT-5.6 LunaOpenAI20.0
39Nemotron 3 UltraNVIDIA20.0
40Qwen3.5 FlashAlibaba20.0
41Muse Spark 1.3Meta19.6
42GPT-5.6 TerraOpenAI19.0
43Kimi K3Kimi19.0
44Qwen3.6 Max PreviewAlibaba19.0
45GLM-5.2Zhipu18.0
46GLM-5.2Zhipu18.0
47GPT-5.5OpenAI18.0
48Qwen3.5 397B-A17BAlibaba18.0
49Qwen3.6 FlashAlibaba18.0
50GPT 5.1OpenAI17.5
51Claude Sonnet 4.5Anthropic17.0
52DeepSeek v4DeepSeek17.0
53GPT-5.6 LunaOpenAI17.0
54Mistral Large 4Mistral17.0
55Qwen3.5-122B-A10BAlibaba17.0
56Qwen3.5 PlusAlibaba17.0
57Qwen3.7 PlusAlibaba17.0
58Qwen3.7 PlusAlibaba17.0
59Claude Opus 4.6Anthropic16.0
60Claude Sonnet 4.6Anthropic16.0
61Claude Sonnet 5Anthropic16.0
62GPT-5.4OpenAI16.0
63Qwen3.5 FlashAlibaba16.0
64Qwen3.5 PlusAlibaba16.0
65Gemini 3.5 Flash-LiteGoogle15.5
66GPT 5OpenAI15.5
67Claude Opus 4.6Anthropic15.0
68GPT-5.1OpenAI15.0
69Kimi K2.6Kimi15.0
70Qwen3.7 FlashAlibaba15.0
71Claude Sonnet 4.6Anthropic14.0
72GPT-5.2OpenAI14.0
73GPT-5.6 TerraOpenAI14.0
74Qwen3.7 FlashAlibaba14.0
75Claude Opus 4.7Anthropic13.0
76GPT-4OpenAI12.0
77GPT-4o miniOpenAI12.0
78Qwen3.6 FlashAlibaba12.0
79Qwen3.6 PlusAlibaba12.0
80GPT-5.4 miniOpenAI11.0
81GPT-5.4 nanoOpenAI9.0
82Qwen3-235B-A22BAlibaba9.0
83GLM-5.3-FlashZhipu8.0
84GPT 5 nanoOpenAI8.0
85MiniMax M3MiniMax8.0
86GPT-5.4 miniOpenAI7.5
87GPT 4.1 miniOpenAI7.0
88GPT-6 LunaOpenAI7.0
89MiniMax M3MiniMax7.0
90O3 MiniOpenAI7.0
91Qwen3.6 27BAlibaba7.0
92GPT 5 miniOpenAI6.0
93Inkling SmallThinking Machines6.0
94GPT-5.4 nanoOpenAI5.0
95o4-miniOpenAI5.0
96Qwen3 MaxAlibaba5.0
97GPT-3.5 TurboOpenAI3.0
98gpt-oss-120bOpenAI1.0

The benchmark's own page