Skip to content

Benchmark

ARC-AGI

Scores are percentages.

Models measured
102
Scores recorded
295
Sources
2
IQ weight
2.26

Score distribution

64.1%
Mean
75%
Median
31.2%
Std dev
14.5%
10th percentile
97%
90th percentile
0%
Min
0%98.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.26 for chat models, measured over 102 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51Claude Fable 5.2low· Epoch AI Benchmarking Hub data94.5%
52DeepSeek V4.1 Flashmax· Epoch AI Benchmarking Hub data94.5%
53DeepSeek V4.1 Flash 1017high· Epoch AI Benchmarking Hub data94.5%
54GPT 5.5high· Epoch AI Benchmarking Hub data94.5%
55GPT-5.4 Pro· Epoch AI Benchmarking Hub data94.5%
56GPT-6.2 Solmedium· Epoch AI Benchmarking Hub data94.5%
57Kimi K3max· Epoch AI Benchmarking Hub data94.5%
58Claude Opus 4.6120k· Epoch AI Benchmarking Hub data94%
59GPT-5.6 Terraxhigh· Epoch AI Benchmarking Hub data94%
60GPT 5.4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim93.7%
61GPT 5.4xhigh· Epoch AI Benchmarking Hub data93.7%
62Claude Opus 4.7high· Epoch AI Benchmarking Hub data93.5%
63DeepSeek V4.1 Flash 1017max· Epoch AI Benchmarking Hub data93.5%
64GPT-6.1 Sollow· Epoch AI Benchmarking Hub data93.5%
65GPT 5.4high· Epoch AI Benchmarking Hub data92.7%
66GPT-6 Solxhigh· Epoch AI Benchmarking Hub data92.7%
67Claude Fable 5medium· Epoch AI Benchmarking Hub data92.5%
68Claude Opus 4.8max· Epoch AI Benchmarking Hub data92.5%
69Gemini 3.5 Flashhigh· Epoch AI Benchmarking Hub data92.5%
70GPT-5.6 Solmedium· Epoch AI Benchmarking Hub data92.5%
71GPT 5.5medium· Epoch AI Benchmarking Hub data92.2%
72Claude Opus 4.7max· Epoch AI Benchmarking Hub data92%
73Claude Opus 4.8high· Epoch AI Benchmarking Hub data92%
74Claude Sonnet 5.5high· Epoch AI Benchmarking Hub data92%
75GPT-5.6 Terrahigh· Epoch AI Benchmarking Hub data92%
76Claude Opus 4.8medium· Epoch AI Benchmarking Hub data91.5%
77DeepSeek V4.1 Flash 1017low· Epoch AI Benchmarking Hub data91.5%
78Gemini 3.6 Flashhigh· Epoch AI Benchmarking Hub data91.2%
79Gemini 3.7 Flashmedium· Epoch AI Benchmarking Hub data91.2%
80Claude Opus 4.7low· Epoch AI Benchmarking Hub data91%
81Claude Opus 4.7medium· Epoch AI Benchmarking Hub data91%
82GLM 5.3 Flashmax· Epoch AI Benchmarking Hub data91%
83GPT-6 Solhigh· Epoch AI Benchmarking Hub data91%
84Claude Fable 5low· Epoch AI Benchmarking Hub data90.5%
85DeepSeek V4 Pro 0813low· Epoch AI Benchmarking Hub data90.5%
86DeepSeek V4.1 Flashlow· Epoch AI Benchmarking Hub data90.5%
87Gemini 3.8 Flashlow· Epoch AI Benchmarking Hub data90.5%
88GPT-5.2 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim90.5%
89GPT-5.2 Proxhigh· Epoch AI Benchmarking Hub data90.5%
90Grok 4.7high· Epoch AI Benchmarking Hub data90.2%
91Claude Fable 5.1low· Epoch AI Benchmarking Hub data90%
92DeepSeek V4 Pro 0813max· Epoch AI Benchmarking Hub data90%
93Grok 4.7medium· Epoch AI Benchmarking Hub data90%
94Grok 4.20· Epoch AI Benchmarking Hub data89.5%
95Grok 4.7xhigh· Epoch AI Benchmarking Hub data89.5%
96Claude Haiku 5.5max· Epoch AI Benchmarking Hub data89%
97DeepSeek V4 Flash 0731max· Epoch AI Benchmarking Hub data89%
98Muse Spark Open 816B A42Bxhigh· Epoch AI Benchmarking Hub data89%
99Claude Opus 5.5low· Epoch AI Benchmarking Hub data88.5%
100DeepSeek V4.1 Flashhigh· Epoch AI Benchmarking Hub data88.5%

Showing 51–100 of 295 results.

Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · Data by LLM Stats