ARC-AGI-1 — leaderboard
ARC-AGI-1 evaluates model capability on agentic tasks from the linked upstream source with Score as the primary reported metric.
Metric: Accuracy (%). Source: arcprize.org. Status: saturated. 164 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 98 |
| 2 | GPT-5.6 Sol (xHigh) | 97.5 |
| 3 | GPT-5.6 Sol (High) | 97 |
| 4 | GPT-5.6 Sol (Max) | 96.5 |
| 5 | GPT-5.6 Terra (Max) | 96.5 |
| 6 | Gemini 3 Deep Think | 96 |
| 7 | GPT-5.5 (xHigh) | 95 |
| 8 | GPT-5.5 Pro (xHigh) | 95 |
| 9 | GPT-5.5 (High) | 94.5 |
| 10 | GPT-5.4 Pro (xHigh) | 94.5 |
| 11 | GPT-5.6 Terra (xHigh) | 94 |
| 12 | GPT-5.4 (xHigh) | 93.67 |
| 13 | GPT-5.4 (High) | 92.67 |
| 14 | Gemini 3.5 Flash (High) | 92.5 |
| 15 | Claude Opus 4.8 (Max) | 92.5 |
Interactive version: theaggregate.ai/benchmark?slug=arc-agi-1 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.