Benchmark · KnowledgeWide open

Mystery Game Puzzles

Updated 2026-09-28
Models tested
53
Top score
82.4
GPT-6 Astra
Median
17.4
min 0.9
Top-5 spread
σ 10.5
wide open

Best score over time · one chart, every benchmark

MYSTERY GAME PUZZLES50 MODELS · FRONTIER RUNNING MAX0255075100SCORE ↑Apr 25Aug 25Jan 26May 26Sep 26RELEASE DATE →benchgecko.ai/benchmark/mystery-game-puzzles · frontier
Frontier on Mystery Game Puzzles rose from 21.8 to 82.4 in 17 months · +60.6 points · latest leader GPT-6 Astra from OpenAI.
Pink dots = frontier records · 7 totalClick to open model page

53 models tested · sorted by score

#ModelScore
1OpenAI logoGPT-6 Astra82.4
2Anthropic logoClaude Opus 5.568.0
3Anthropic logoClaude Sonnet 5.561.5
4Anthropic logoClaude Opus 554.8
5Anthropic logoClaude Fable 5.153.7
6OpenAI logoGPT-5.6 Sol53.7
7Anthropic logoClaude Fable 547.1
8Google DeepMind logoGemini 3.8 Flash41.6
9DeepSeek logoDeepSeek V4 Pro 081337.2
10Alibaba Qwen logoQwen3.8 Max (0902)31.7
11Google DeepMind logoGemini 3.7 Flash30.6
12OpenAI logoGPT-5.430.6
13Anthropic logoClaude Opus 4.829.5
14Anthropic logoClaude Sonnet 528.4
15OpenAI logoGPT-5.6 Terra28.4
16DeepSeek logoDeepSeek V4 Flash 073127.3
17Google DeepMind logoGemini 3.1 Pro Preview27.3
18xAI logoGrok 4.627.3
19z-ai logoGLM 5.326.2
20Google DeepMind logoGemini 3.5 Flash25.1
21Alibaba Qwen logoQwen3.7 Max25.1
22Google DeepMind logoGemini 3.6 Flash22.9
23OpenAI logoo321.8
24Anthropic logoClaude Opus 4.720.7
25Google DeepMind logoGemini 3 Flash Preview18.5
26moonshotai logoKimi K318.5
27Anthropic logoClaude Opus 4.617.4
28OpenAI logoGPT-515.2
29OpenAI logoGPT-5.215.2
30Anthropic logoClaude Opus 4.514.1
31Alibaba Qwen logoQwen3.6 35B A3B14.1
32Anthropic logoClaude Opus 4.113.0
33OpenAI logoGPT-5.6 Luna13.0
34NVIDIA logoNemotron 3 Ultra11.9
35Alibaba Qwen logoQwen3.5-Flash11.9
36Google DeepMind logoGemini 3.5 Flash Lite10.8
37z-ai logoGLM 5.210.8
38OpenAI logoGPT-5.110.8
39Alibaba Qwen logoQwen3.6 Max Preview10.8
40moonshotai logoKimi K2.69.7
41Alibaba Qwen logoQwen3.5 397B A17B9.7
42Alibaba Qwen logoQwen3.6 Flash9.7
43Anthropic logoClaude Sonnet 4.58.6
44DeepSeek logoDeepSeek V4 Pro8.6
45Alibaba Qwen logoQwen 3.5 Plus (hosted 397B-A17B)8.6
46Alibaba Qwen logoQwen3.7 Plus8.6
47Anthropic logoClaude Sonnet 4.67.5
48Alibaba Qwen logoQwen3.7 Flash6.4
49OpenAI logoGPT-4 Turbo3.1
50OpenAI logoGPT-4o-mini3.1
51Alibaba Qwen logoQwen3.6 Plus3.1
52OpenAI logoGPT-5.4 Mini2.0
53OpenAI logoGPT-5 Mini0.9

Same category · related evaluations