benchgap
General

Mystery Game Puzzles leaderboard

As of 2026-11-18, the highest measured score on Mystery Game Puzzles is 95.0% by GPT-6.2 Astra. 70 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-6.2 Astra95.0%measured
2Claude Fable 5.290.0%measured
3GPT-6.2 Sol87.5%estimated ± 6.4 pp, low confidence
4GPT-6 Cyber86.3%estimated ± 6.4 pp, low confidence
5GPT-6 Astra84.0%measured
6GPT-6.1 Sol80.0%measured
7Claude Mythos 5.277.5%estimated ± 7.5 pp, medium confidence
8GPT-6.1 Terra74.0%measured
9Gemini 4 Carbon73.0%measured
10Claude Opus 5.571.0%measured
11Claude Sonnet 5.565.0%measured
12Gemini 4 Argon64.0%measured
13Hy463.6%estimated ± 12.6 pp, low confidence
14ERNIE 5.5 Flash Preview61.0%estimated ± 12.6 pp, low confidence
15Claude Opus 559.0%measured
16GPT-5.5 Pro58.7%estimated ± 11.3 pp, low confidence
17Claude Fable 5.158.0%measured
18GPT-5.6 Sol58.0%measured
19GPT-5.556.0%measured
20GPT-6 Sol56.0%measured
21Claude Mythos 554.8%estimated ± 7.5 pp, medium confidence
22GPT-5.4 Pro53.7%estimated ± 11.3 pp, low confidence
23Claude Fable 552.0%measured
24Muse Wonder51.0%measured
25Claude Opus 4.6 (Adaptive)49.1%estimated ± 14.6 pp, low confidence
26DeepSeek V4.1 Flash 101749.0%measured
27Gemini 3.8 Flash47.0%measured
28Grok 4.20 Multi-agent45.3%estimated ± 14.6 pp, low confidence
29Claude Mythos Preview45.1%estimated ± 7.5 pp, medium confidence
30Grok 4.1 Fast (Reasoning)42.0%estimated ± 14.6 pp, low confidence
31Kimi K3.1 Pro42.0%measured
32Qwen3.8 Max38.0%measured
33Gemini 3.7 Flash37.0%measured
34GPT-5.437.0%measured
35Claude Opus 4.836.0%measured
36Claude Sonnet 535.0%measured
37GPT-5.6 Terra35.0%measured
38DeepSeek V4 Flash 073134.0%measured
39Gemini 3.1 Pro34.0%measured
40Grok 4.634.0%measured
41GLM-5.333.0%measured
42Gemini 3.5 Flash32.0%measured
43Qwen3.7 Max32.0%measured
44MiniMax M3.1 Flash31.8%estimated ± 13.3 pp, low confidence
45GLM-5.130.8%estimated ± 9.8 pp, medium confidence
46Muse Spark Open30.4%estimated ± 13.3 pp, low confidence
47Hy4 preview30.2%estimated ± 13.3 pp, low confidence
48Claude Haiku 5.530.0%measured
49Gemini 3.6 Flash30.0%measured
50DeepSeek V4 Pro29.0%estimated ± 13.3 pp, low confidence
51Grok 4.729.0%measured
52o329.0%measured
53Ling 3.1 Flash28.6%estimated ± 13.3 pp, low confidence
54Claude Opus 4.728.0%measured
55Grok 4.528.0%estimated ± 10.7 pp, low confidence
56Hy327.0%estimated ± 9.8 pp, medium confidence
57MiMo-V2.6-Pro27.0%estimated ± 7.5 pp, medium confidence
58Kimi K326.0%measured
59Qwen3.8-27B25.9%estimated ± 12.4 pp, low confidence
60Step 5 Preview25.3%estimated ± 13.3 pp, low confidence
61Claude Opus 4.625.0%measured
62Muse Spark 1.325.0%measured
63Muse Spark 1.123.5%estimated ± 11.3 pp, low confidence
64GPT-523.0%measured
65GPT-5.223.0%measured
66Claude Opus 4.522.0%measured
67Qwen3.6-35B-A3B22.0%measured
68GPT-5.6 Luna21.0%measured
69Grok 4.2020.2%estimated ± 11.3 pp, low confidence
70Gemini 3 Flash20.0%measured
71Nemotron 3 Ultra20.0%measured
72Qwen3.5 Flash20.0%measured
73Beam 1.0 501B-A23B19.1%estimated ± 13.3 pp, low confidence
74Muse Spark 1.219.1%estimated ± 11.3 pp, low confidence
75Claude Sonnet 4.5 Thinking19.1%estimated ± 4.7 pp, medium confidence
76Gemini 3.5 Flash-Lite19.0%measured
77Qwen 3.6 Max (preview)19.0%measured
78Seed 2.1 Turbo18.6%estimated ± 11.3 pp, low confidence
79SWE-2 Flash18.0%estimated ± 7.5 pp, medium confidence
80Kimi K2.618.0%measured
81Qwen3.5 397B18.0%measured
82Gemma 4 31B17.2%estimated ± 13.3 pp, low confidence
83Claude Sonnet 4.517.0%measured
84Mistral Large 417.0%measured
85Qwen3.5-122B-A10B17.0%measured
86Qwen3.5 Plus17.0%measured
87Qwen3.7 Plus17.0%measured
88Gemini 3 Pro16.9%estimated ± 2.2 pp, medium confidence
89GPT-5.2-Codex16.9%estimated ± 2.2 pp, medium confidence
90Inkling16.4%estimated ± 11.3 pp, low confidence
91Seed 2.1 Pro16.4%estimated ± 11.3 pp, low confidence
92Claude Sonnet 4.616.0%measured
93GPT-5.116.0%measured
94Claude Opus 4.5 Thinking15.3%estimated ± 14.6 pp, low confidence
95GLM-4.715.1%estimated ± 2.2 pp, low confidence
96GLM-5.215.0%measured
97Qwen3.7 Flash15.0%measured
98Grok 4.314.7%estimated ± 9.8 pp, medium confidence
99Claude Haiku 4.5 Thinking14.5%estimated ± 4.7 pp, medium confidence
100Qwen3.5-27B14.5%estimated ± 14.6 pp, low confidence
101Kimi K2.7 Code14.2%estimated ± 11.3 pp, low confidence
102DeepSeek V3.214.1%estimated ± 2.2 pp, low confidence
103GPT-5.1-Codex-Max14.1%estimated ± 2.2 pp, low confidence
104Step 3.7 Flash13.8%estimated ± 14.6 pp, low confidence
105MiMo-V2.5-Pro13.6%estimated ± 14.6 pp, low confidence
106MiMo-V2-Pro13.5%estimated ± 14.6 pp, low confidence
107Hy3 Preview13.5%estimated ± 14.6 pp, low confidence
108Trinity-Large-Thinking13.5%estimated ± 14.6 pp, low confidence
109Mistral Medium 3.5 128B13.5%estimated ± 14.6 pp, low confidence
110Llama 4 Maverick13.5%estimated ± 14.6 pp, low confidence
111Mistral Large 313.5%estimated ± 14.6 pp, low confidence
112o113.2%estimated ± 11.3 pp, low confidence
113GPT-4o13.0%estimated ± 11.3 pp, low confidence
114MiniMax M2.712.8%estimated ± 9.8 pp, low confidence
115MiMo-V2.6-Flash12.4%estimated ± 7.5 pp, low confidence
116Claude 3 Opus12.2%estimated ± 11.3 pp, low confidence
117GPT-4o mini12.0%measured
118Qwen3.6 Plus12.0%measured
119Gemini 3.1 Flash-Lite11.8%estimated ± 9.8 pp, low confidence
120GPT-4.1 nano11.7%estimated ± 11.3 pp, low confidence
121Gemma 4 26B A4B11.4%estimated ± 13.3 pp, low confidence
122DeepSeek V3.110.9%estimated ± 4.7 pp, medium confidence
123Grok Code Fast 110.9%estimated ± 4.7 pp, medium confidence
124Claude Haiku 4.58.7%estimated ± 2.2 pp, low confidence
125GLM-5.3-Flash8.0%measured
126MiniMax M38.0%measured
127Grok 4 Fast (Reasoning)7.3%estimated ± 4.7 pp, medium confidence
128MAI-Thinking-17.1%estimated ± 13.3 pp, low confidence
129GPT-4.1 mini7.0%measured
130GPT-5.4 mini7.0%measured
131GPT-6 Luna7.0%measured
132o3-mini7.0%measured
133Qwen3.6-27B7.0%measured
134Grok 46.9%estimated ± 2.2 pp, low confidence
135Gemini 2.5 Pro6.4%estimated ± 4.7 pp, medium confidence
136GPT-4.16.4%estimated ± 4.7 pp, medium confidence
137Inkling-Small6.0%measured
138GPT-5.4 nano5.0%measured
139o4-mini5.0%measured
140Qwen3 Max5.0%measured
141Grok 4.1 Fast4.9%estimated ± 2.2 pp, low confidence
142GLM-4.54.6%estimated ± 4.7 pp, medium confidence
143Gemini 2.5 Flash1.0%estimated ± 4.7 pp, low confidence
144GPT-OSS 120B0.0%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General