Skip to content

Benchmark

ARC-AGI 2

Scores are percentages. Higher is better.

Models measured
98
Scores recorded
318
Sources
3
IQ weight
1.29

Score distribution

42.4%
Mean
42.5%
Median
35%
Std dev
0.9%
10th percentile
90%
90th percentile
0%
Min
0%95.8%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.29 for chat models, measured over 98 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
101Claude Opus 4.7medium· Epoch AI Benchmarking Hub data67.5%
102DeepSeek V4.1 Flashhigh· Epoch AI Benchmarking Hub data67.5%
103GPT 5.4high· Epoch AI Benchmarking Hub data67.5%
104Claude Sonnet 5.5high· Epoch AI Benchmarking Hub data67.4%
105Grok 4.6xhigh· Epoch AI Benchmarking Hub data67.1%
106GPT-5.6 Solmedium· Epoch AI Benchmarking Hub data67.1%
107GPT-5.6 Terrahigh· Epoch AI Benchmarking Hub data67.1%
108GLM 5.3 Flashmax· Epoch AI Benchmarking Hub data65.8%
109Grok 4.20· Epoch AI Benchmarking Hub data65.1%
110Grok 4.6high· Epoch AI Benchmarking Hub data65.1%
111Gemini 3.7 Flashmedium· Epoch AI Benchmarking Hub data63.8%
112Doubao Seed 2.1 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim62.5%
113Claude Opus 4.8low· Epoch AI Benchmarking Hub data62.2%
114Claude Opus 4.7low· Epoch AI Benchmarking Hub data62.1%
115DeepSeek V4 Flash 0731max· Epoch AI Benchmarking Hub data61.4%
116Grok 4.7xhigh· Epoch AI Benchmarking Hub data61.4%
117Doubao Seed 2.1 Turbo· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim61.3%
118Claude Haiku 5.5xhigh· Epoch AI Benchmarking Hub data61.3%
119DeepSeek V4 Pro 0813max· Epoch AI Benchmarking Hub data61.3%
120Grok 4.6medium· Epoch AI Benchmarking Hub data61.3%
121DeepSeek V4.1 Flashlow· Epoch AI Benchmarking Hub data60.6%
122Claude Sonnet 4.6high· Epoch AI Benchmarking Hub data60.4%
123Claude Sonnet 4.6private set · 120K thinking tokens · high effort· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.60.4%
124Gemini 3.6 Flashhigh· Epoch AI Benchmarking Hub data60.4%
125Kimi K3max· Epoch AI Benchmarking Hub data60.4%
126DeepSeek V4 Pro 0813high· Epoch AI Benchmarking Hub data59.7%
127GPT-6 Astranone· Epoch AI Benchmarking Hub data59.6%
128GPT-5.6 Lunamax· Epoch AI Benchmarking Hub data59.5%
129GPT-6 Lunamax· Epoch AI Benchmarking Hub data59.3%
130Grok 4.7medium· Epoch AI Benchmarking Hub data58.8%
131Grok 4.7high· Epoch AI Benchmarking Hub data58.3%
132Claude Sonnet 4.6max· Epoch AI Benchmarking Hub data58.3%
133Claude Sonnet 4.6Gemini 3.5 Flash Model Card · May 2026· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.58.3%
134Claude Sonnet 4.6Verified · adaptive thinking · max effort· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.58.3%
135GPT-6.2 Sollow· Epoch AI Benchmarking Hub data57.9%
136GPT-6 Solmedium· Epoch AI Benchmarking Hub data57.8%
137Muse Spark Open 816B A42Bxhigh· Epoch AI Benchmarking Hub data57.5%
138DeepSeek V4 Pro 0813low· Epoch AI Benchmarking Hub data56.3%
139DeepSeek V4 Flash 0731high· Epoch AI Benchmarking Hub data56%
140GPT 5.4medium· Epoch AI Benchmarking Hub data55.4%
141Kimi K3high· Epoch AI Benchmarking Hub data55%
142GPT 5.2Opus 4.6 System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials unless noted · Verified· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.54.2%
143GPT-5.2 ProVerified · high effort· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.54.2%
144GPT-5.2 Prohigh· Epoch AI Benchmarking Hub data54.2%
145Gemini 3.7 Flashlow· Epoch AI Benchmarking Hub data52.9%
146GPT 5.2xhigh· Epoch AI Benchmarking Hub data52.9%
147GPT 5.2Verified· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.9%
148GPT 5.2Verified · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.52.9%
149Grok 4.5high· Epoch AI Benchmarking Hub data52.6%
150Grok 4.5medium· Epoch AI Benchmarking Hub data52.6%

Showing 101–150 of 318 results.

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · Data by LLM Stats