benchgap
General

Chess Puzzles leaderboard

As of 2026-11-19, the highest measured score on Chess Puzzles is 79.0% by Claude Fable 5.2. 54 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Fable 5.279.0%measured
2Gemini 4 Carbon78.0%measured
3GPT-6.2 Astra78.0%measured
4GPT-6.2 Sol76.8%estimated ± 5.3 pp, low confidence
5GPT-6 Cyber75.0%estimated ± 5.3 pp, low confidence
6GPT-6 Astra74.0%measured
7Claude Mythos 5.268.8%estimated ± 8.0 pp, medium confidence
8GPT-6.1 Sol68.0%measured
9Claude Opus 5.567.0%measured
10Gemini 3.7 Flash66.0%measured
11GPT-6.1 Terra65.9%estimated ± 4.7 pp, high confidence
12GPT-5.563.0%measured
13GPT-5.5 Pro62.0%measured
14Gemini 4 Hydrogen60.0%measured
15Claude Sonnet 5.559.0%measured
16Gemini 4 Argon59.0%measured
17GPT-5.4 Pro59.0%measured
18GPT-5.6 Sol59.0%measured
19Hy458.4%estimated ± 8.0 pp, low confidence
20Gemini 3.8 Flash58.0%measured
21ERNIE 5.5 Flash Preview56.6%estimated ± 8.0 pp, low confidence
22GPT-6 Sol56.0%measured
23Gemini 3 Pro55.2%estimated ± 12.3 pp, low confidence
24Claude Haiku 5.555.0%measured
25Claude Mythos 554.4%estimated ± 8.0 pp, medium confidence
26Muse Wonder53.2%estimated ± 4.7 pp, high confidence
27Claude Opus 4.852.0%measured
28Claude Opus 4.6 (Adaptive)51.8%estimated ± 12.3 pp, low confidence
29Gemini 3.1 Pro51.0%measured
30Claude Mythos Preview48.5%estimated ± 8.0 pp, medium confidence
31Grok 4.20 Multi-agent48.3%estimated ± 12.3 pp, low confidence
32Qwen3.8-27B48.2%estimated ± 10.8 pp, low confidence
33Claude Fable 5.148.0%measured
34GPT-5.248.0%measured
35GPT-5.448.0%measured
36Claude Fable 547.0%measured
37GPT-5.6 Terra47.0%measured
38DeepSeek V4.1 Flash 101746.0%measured
39Qwen3.8 Max46.0%measured
40Grok 4.1 Fast (Reasoning)45.3%estimated ± 12.3 pp, low confidence
41Claude Sonnet 545.0%measured
42Gemini 3.5 Flash45.0%measured
43GPT-5.4 nano45.0%measured
44Kimi K3.1 Pro45.0%measured
45MiniMax M3.1 Flash43.8%estimated ± 10.8 pp, low confidence
46Claude Opus 543.0%measured
47GPT-5.6 Luna43.0%measured
48Muse Spark Open42.8%estimated ± 10.8 pp, low confidence
49Hy4 preview42.6%estimated ± 10.8 pp, low confidence
50GPT-5.142.0%measured
51DeepSeek V4 Pro41.8%estimated ± 10.8 pp, low confidence
52Ling 3.1 Flash41.5%estimated ± 10.8 pp, low confidence
53MiMo-V2.6-Pro41.1%estimated ± 8.0 pp, medium confidence
54Claude Opus 4.741.0%measured
55Grok 4.741.0%measured
56Gemini 3.6 Flash40.0%measured
57Gemini 3 Flash40.0%measured
58SWE-2 Flash39.6%estimated ± 8.0 pp, medium confidence
59MiMo-V2.6-Flash39.3%estimated ± 8.0 pp, low confidence
60Step 5 Preview38.8%estimated ± 10.8 pp, low confidence
61Qwen3.6-35B-A3B37.2%estimated ± 4.7 pp, high confidence
62GLM-5.337.0%measured
63Muse Spark 1.337.0%measured
64Hy337.0%estimated ± 11.1 pp, low confidence
65Nemotron 3 Ultra36.1%estimated ± 4.7 pp, medium confidence
66GPT-536.0%measured
67GPT-6 Luna36.0%measured
68Kimi K336.0%measured
69o336.0%measured
70Qwen 3.6 Max (preview)35.5%estimated ± 4.7 pp, medium confidence
71Grok 4.535.0%measured
72Qwen3.5 397B35.0%estimated ± 4.7 pp, medium confidence
73Qwen3.5-122B-A10B34.4%estimated ± 4.7 pp, medium confidence
74Qwen3.7 Plus34.4%estimated ± 4.7 pp, medium confidence
75Muse Spark 1.134.0%measured
76Qwen3.7 Flash33.3%estimated ± 4.7 pp, medium confidence
77Grok 4.633.0%measured
78Mistral Large 433.0%measured
79Beam 1.0 501B-A23B32.9%estimated ± 10.8 pp, low confidence
80DeepSeek V4 Flash 073132.0%measured
81Gemma 4 31B30.8%estimated ± 10.8 pp, low confidence
82Grok 4.2029.0%measured
83Qwen3.6-27B28.9%estimated ± 4.7 pp, medium confidence
84Inkling-Small28.3%estimated ± 4.7 pp, medium confidence
85Qwen3 Max27.8%estimated ± 4.7 pp, medium confidence
86Muse Spark 1.227.0%measured
87Qwen3.7 Max27.0%measured
88Seed 2.1 Turbo26.0%measured
89GPT-OSS 120B25.0%estimated ± 4.7 pp, medium confidence
90Gemma 4 26B A4B23.1%estimated ± 10.8 pp, low confidence
91Claude Sonnet 4.623.0%measured
92GLM-5.222.0%measured
93Claude Opus 4.621.0%measured
94GLM-4.721.0%measured
95GLM-5.3-Flash21.0%measured
96Inkling21.0%measured
97Seed 2.1 Pro21.0%measured
98Claude Opus 4.5 Thinking20.9%estimated ± 12.3 pp, low confidence
99Qwen3.5-27B20.2%estimated ± 12.3 pp, low confidence
100GLM-5.120.0%measured
101Kimi K2.620.0%measured
102Step 3.7 Flash19.6%estimated ± 12.3 pp, low confidence
103Claude Sonnet 4.5 Thinking19.5%estimated ± 12.3 pp, low confidence
104MiMo-V2.5-Pro19.4%estimated ± 12.3 pp, low confidence
105MiMo-V2-Pro19.3%estimated ± 12.3 pp, low confidence
106MiniMax M2.719.3%estimated ± 12.3 pp, low confidence
107Hy3 Preview19.3%estimated ± 12.3 pp, low confidence
108Trinity-Large-Thinking19.3%estimated ± 12.3 pp, low confidence
109Mistral Medium 3.5 128B19.3%estimated ± 12.3 pp, low confidence
110Grok 4.1 Fast19.3%estimated ± 12.3 pp, low confidence
111DeepSeek V3.219.3%estimated ± 12.3 pp, low confidence
112Gemini 3.1 Flash-Lite19.3%estimated ± 12.3 pp, low confidence
113Llama 4 Maverick19.3%estimated ± 12.3 pp, low confidence
114Mistral Large 319.3%estimated ± 12.3 pp, low confidence
115Grok 4.318.0%measured
116GPT-5.4 mini17.0%measured
117MiniMax M317.0%measured
118MAI-Thinking-116.0%estimated ± 10.8 pp, low confidence
119Gemini 3.5 Flash-Lite16.0%measured
120Qwen3.6 Plus16.0%measured
121o3-mini15.0%measured
122Claude Opus 4.514.0%measured
123Kimi K2.7 Code14.0%measured
124Qwen3.5 Flash14.0%measured
125o4-mini13.0%measured
126GPT-4.111.0%measured
127Claude Sonnet 4.510.0%measured
128o110.0%measured
129GPT-4o9.0%measured
130Qwen3.5 Plus9.0%measured
131Claude Haiku 4.58.0%measured
132Claude 3 Opus4.0%measured
133GPT-4o mini3.0%measured
134GPT-4.1 mini1.0%measured
135GPT-4.1 nano0.0%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General