ARC-AGI-2
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
The Frontier
Best score over time · one chart, every benchmark
Full rankings
91 models tested · sorted by score
What it measures · how it is scored · sources · as of 2026-11-22
- What it measures · ARC-AGI 2, harder sequel to ARC. More complex abstract reasoning patterns that test generalization ability beyond training data.
- How it is scored · Scored as a percentage · higher is better · maximum 100
- Released · 2025-03-24
- Scores from · Epoch AI Benchmarking Hub (91 scores)
Score distribution
Where models cluster
Correlated benchmarks
Pearson r · original research
Benchmarks that track with ARC-AGI-2
Pearson correlation across models scored on both benchmarks. Closer to 1 = strongly predictive.
Frequently asked
About ARC-AGI-2
What does ARC-AGI-2 measure?
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data. 91 AI models have been tested on it. Scores range from 0.1 to 95.8 out of 100.
Which model leads on ARC-AGI-2?
GPT-6.2 Astra from OpenAI leads ARC-AGI-2 with a score of 95.8. The median score across 91 tested models is 33.6.
Is ARC-AGI-2 saturated?
Yes · the top model on ARC-AGI-2 has reached 95.8 out of 100, within 5% of the theoretical ceiling. This benchmark is approaching saturation and may be replaced by a harder successor.
Does ARC-AGI-2 predict performance on other benchmarks?
Yes · ARC-AGI-2 scores correlate 0.95 with Professional Reasoning · Legal across 5 shared models. Models that do well on ARC-AGI-2 tend to do well on Professional Reasoning · Legal.
How often is ARC-AGI-2 data refreshed?
BenchGecko pulls updates daily. New model scores on ARC-AGI-2 appear as soon as they are published by Epoch AI or the model provider.
- Category
- Reasoning
- Max score
- 100
- Models
- 91
- Updated
- 2026-11-17
Top on ARC-AGI-2
GPT-6.2 Astra · 95.8GPT-6 Astra · 95.0Gemini 4 Carbon · 94.2GPT-6.1 Sol · 94.2Claude Fable 5.2 · 93.8More reasoning benchmarks
Same category · related evaluations