modelbenchmark.io
All benchmarks

Chess Puzzles

Reasoning · questions rotate

Tactical chess positions with one correct continuation.

How far to trust it

A narrow skill, and not one most buyers need. It is in the composite because it discriminates well, not because it is representative.

Measured

Models scored
99
Spread between models
19.8
standard deviation, points
Measurement noise
4.4
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 AstraOpenAI76.0
2GPT-6 AstraOpenAI74.0
3GPT-5.6 SolOpenAI68.0
4GPT-6.1 SolOpenAI68.0
5Claude Opus 5.5Anthropic67.0
6Gemini 3.7 FlashGoogle66.0
7GPT-5.5 ProOpenAI62.0
8GPT-5.2 ProOpenAI60.0
9Claude Sonnet 5.5Anthropic59.0
10Gemini 4 ArgonGoogle59.0
11GPT-5.4 ProOpenAI59.0
12Gemini 3.8 FlashGoogle58.0
13GPT-6 SolOpenAI56.0
14Gemini 3.1 Pro PreviewGoogle51.0
15Claude Fable 5.1Anthropic48.0
16DeepSeek V4.1 FlashDeepSeek48.0
17GPT-5 ProOpenAI48.0
18GPT-5.5OpenAI46.5
19DeepSeek V4 Pro 0813DeepSeek46.0
20Qwen3.8 MaxAlibaba46.0
21Claude Fable 5Anthropic45.0
22GPT-5.6 SolOpenAI45.0
23Gemini 3.5 FlashGoogle42.0
24Grok 4.7xAI41.0
25Claude Opus 4.8Anthropic40.5
26GPT 5.2OpenAI40.5
27Gemini 3.6 FlashGoogle40.0
28Claude Haiku 5.5Anthropic39.5
29Claude Sonnet 5Anthropic39.5
30GPT-5.4OpenAI39.5
31Gemini 3 Flash PreviewGoogle39.0
32GLM-5.3Zhipu37.0
33GPT-5.6 TerraOpenAI37.0
34Qwen3.8 Max (0902)Alibaba37.0
35o3OpenAI36.3
36Claude Opus 4.7Anthropic36.0
37GPT-5.4 nanoOpenAI36.0
38GPT-6 LunaOpenAI36.0
39Kimi K3Kimi36.0
40Muse Spark 1.3Meta35.5
41Claude Opus 5Anthropic35.0
42Grok 4.5xAI35.0
43Muse Spark 1.1Meta34.0
44Grok 4.6xAI33.0
45Mistral Large 4Mistral33.0
46DeepSeek V4 Flash 0731DeepSeek32.0
47GPT 5.1OpenAI32.0
48GPT-5.6 LunaOpenAI30.0
49GPT 5OpenAI29.0
50Grok 4.20xAI29.0
51Muse Spark 1.2Meta27.0
52Qwen3.7-MaxAlibaba26.0
53Claude Opus 4.6Anthropic21.0
54GLM-4.7Zhipu21.0
55GLM-5.3-FlashZhipu21.0
56InklingThinking Machines21.0
57GLM-5.1Zhipu20.0
58GLM-5.2Zhipu20.0
59GPT-5.4 miniOpenAI20.0
60Kimi K2.6Kimi20.0
61Claude Opus 4.8Anthropic19.0
62Claude Sonnet 4.6Anthropic19.0
63GLM-5.2Zhipu19.0
64Grok 4.3 BetaxAI18.0
65DeepSeek v4DeepSeek17.0
66o4-miniOpenAI16.7
67Gemini 3.5 Flash-LiteGoogle16.0
68GPT 5 miniOpenAI16.0
69GPT 5 nanoOpenAI15.2
70Kimi K2.7 CodeKimi14.0
71Qwen3.5 FlashAlibaba14.0
72Qwen3.6 PlusAlibaba14.0
73MiniMax M3MiniMax13.0
74Qwen3.6 FlashAlibaba12.0
75O3 MiniOpenAI10.7
76Claude Sonnet 4.5Anthropic10.0
77Claude Opus 4.5Anthropic9.5
78GPT-5.5OpenAI9.0
79o1-2024-12-17OpenAI9.0
80Qwen3.5 PlusAlibaba8.0
81GPT-5.6 SolOpenAI7.0
82MiniMax M3MiniMax7.0
83Claude Haiku 4.5Anthropic6.0
84Claude Sonnet 4.5Anthropic6.0
85GPT 4oOpenAI6.0
86Claude Opus 4.5Anthropic5.0
87GPT-5.6 TerraOpenAI5.0
88DeepSeek v4DeepSeek4.0
89GPT-5.1OpenAI4.0
90Claude 3 OpusAnthropic3.0
91GPT 4.1OpenAI3.0
92GPT-4o miniOpenAI3.0
93GPT-5.4OpenAI3.0
94GPT-5.6 LunaOpenAI3.0
95GPT-5.2OpenAI2.0
96GPT-5.4 miniOpenAI2.0
97GPT-5.4 nanoOpenAI2.0
98GPT 4.1 miniOpenAI1.0
99GPT-4.1 nanoOpenAI0.0

The benchmark's own page