ARC-AGI-3: leaderboard
Third iteration of the Abstraction and Reasoning Corpus from ARC Prize, and the first that is interactive: agents explore novel game environments, work out the goal on the fly and learn as they play, rather than solving static grid puzzles.
Metric: Accuracy (%). Source: arcprize.org. Saturation forecast: Around June 2027. 64 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 62.71 | #6 (GPT-6 Astra) |
| 2 | GPT-6 Astra (xHigh) | 59.34 | #6 (GPT-6 Astra) |
| 3 | GPT-6 Astra (High) | 54.82 | #6 (GPT-6 Astra) |
| 4 | GPT-6.1 Sol (Max) | 52.73 | #7 (GPT-6.1 Sol) |
| 5 | Gemini 4 Argon (Medium) | 46.27 | #8 (Gemini 4 Argon) |
| 6 | GPT-6.1 Sol (xHigh) | 39.93 | #7 (GPT-6.1 Sol) |
| 7 | GPT-6 Astra (Medium) | 38.59 | #6 (GPT-6 Astra) |
| 8 | Claude Opus 5 (High) | 30.16 | #12 (Claude Opus 5) |
| 9 | GPT-6.1 Sol (High) | 26.72 | #7 (GPT-6.1 Sol) |
| 10 | GPT-6 Astra (Low) | 17.45 | #6 (GPT-6 Astra) |
| 11 | GPT-6.1 Sol (Medium) | 10.57 | #7 (GPT-6.1 Sol) |
| 12 | Gemini 3.8 Flash (High) | 10.37 | #21 (Gemini 3.8 Flash) |
| 13 | GPT-5.6 Sol (Max) | 7.78 | #19 (GPT-5.6 Sol) |
| 14 | GPT-5.6 Sol (xHigh) | 6.99 | #19 (GPT-5.6 Sol) |
| 15 | Gemini 3.8 Flash (Low) | 5.99 | #21 (Gemini 3.8 Flash) |
No result here: #3 Claude Opus 5.5, #9 Claude Fable 5.1, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=arc-agi-3 · How It Works · Data refreshed daily, snapshot 2026-10-20.