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
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.
Full rankings
50 models tested · sorted by score
| # | Model | Score |
|---|---|---|
| 1 |  GPT-5.4 Pro | 83.3 |
| 2 |  Gemini 3.1 Pro Preview | 77.1 |
| 3 | GPT-5.4 | 74.0 |
| 4 |  Claude Opus 4.6 | 69.2 |
| 5 | Claude Sonnet 4.6 | 60.4 |
| 6 | GPT-5.2 Pro | 54.2 |
| 7 | GPT-5.2 | 52.9 |
| 8 | Claude Opus 4.5 | 37.6 |
| 9 | Gemini 3 Flash Preview | 33.6 |
| 10 | Gemini 3 Pro | 31.1 |
| 11 | GPT-5 Pro | 18.3 |
| 12 | GPT-5.1 | 17.6 |
| 13 |  Grok 4 | 16.0 |
| 14 | Claude Sonnet 4.5 | 13.6 |
| 15 |  Kimi K2.5 | 11.8 |
| 16 | GPT-5 | 9.9 |
| 17 | Claude Opus 4 | 8.6 |
| 18 | o3 | 6.5 |
| 19 | o4 Mini | 6.1 |
| 20 | Claude Sonnet 4 | 5.9 |
| 21 | Grok 4 Fast | 5.3 |
| 22 | Gemini 2.5 Pro | 4.9 |
| 23 |  GLM 5 | 4.9 |
| 24 |  MiniMax M2.5 | 4.9 |
| 25 | o3 Pro | 4.9 |
| 26 | GPT-5 Mini | 4.4 |
| 27 | Claude Haiku 4.5 | 4.0 |
| 28 |  DeepSeek V3.2 | 4.0 |
| 29 | o3 Mini | 3.0 |
| 30 | GPT-5 Nano | 2.6 |
| 31 | Gemini 2.5 Flash | 2.5 |
| 32 | Gemini 2.0 Flash | 1.3 |
| 33 | R1 | 1.3 |
| 34 |  Qwen3 235B A22B Instruct 2507 | 1.3 |
| 35 | R1 0528 | 1.1 |
| 36 | Claude 3.7 Sonnet | 0.9 |
| 37 | o1-mini | 0.8 |
| 38 | Gemini 1.5 Pro (Feb 2024) | 0.8 |
| 39 | GPT-4.5 | 0.8 |
| 40 | GPT-4.1 | 0.4 |
| 41 | Grok 3 Mini | 0.4 |
| 42 | GPT-4.1 Mini | 0.1 |
| 43 | GPT-4.1 Nano | 0.1 |
| 44 | GPT-4o (2024-11-20) | 0.1 |
| 45 | GPT-4o-mini | 0.1 |
| 46 | GPT-4o-mini (2024-07-18) | 0.1 |
| 47 | Grok 3 | 0.1 |
| 48 |  Llama 4 Maverick | 0.1 |
| 49 | Llama 4 Scout | 0.1 |
| 50 | U Magistral Small 1.1 | 0.1 |
Frequently asked
Pulled from the ARC-AGI-2 dataset · updated daily
What does ARC-AGI-2 measure?
ARC-AGI-2 is a reasoning benchmark in the BenchGecko catalog. 50 AI models have been tested on it. Scores range from 0.1 to 83.3 out of 100.
Which model leads on ARC-AGI-2?
GPT-5.4 Pro from OpenAI leads ARC-AGI-2 with a score of 83.3. The median score across 50 tested models is 4.7.
Is ARC-AGI-2 saturated?
No · the top score is 83.3 out of 100 (83%). There is still meaningful room for improvement on ARC-AGI-2.
Does ARC-AGI-2 predict performance on other benchmarks?
Yes · ARC-AGI-2 scores correlate 0.94 with GSO-Bench across 16 shared models. Models that do well on ARC-AGI-2 tend to do well on GSO-Bench.
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.
More reasoning benchmarks
Same category · related evaluations