ARC-AGI-3 what it measures, how it is scored and who leads it
Reasoning · ARC Prize Foundation · introduced Mar 25, 2026 · Score against human efficiency
About this benchmark
- Task
- Play unfamiliar interactive games with no instructions: explore, find the goal and win as efficiently as a person.
- Dataset
- Game environments that people solve on first contact; a public game set plus held-out games.
- Method
- Scored on skill-acquisition efficiency: how many actions an agent needs compared with people.
- Metric
- Score against human efficiency
- Organization
- ARC Prize Foundation
- Introduced
- Mar 25, 2026
- Official leaderboard
- arcprize.org/arc-agi/3/
Versions
Newest first. New versions are added, never rewritten.
| Version | Date | |
|---|---|---|
| 3Launched, with a technical report and the ARC Prize 2026 track. | Mar 25, 20266 months ago | Launched, with a technical report and the ARC Prize 2026 track. |
| PreviewA small public preview; ARC Prize published 30-day learnings on 2025-08-19. | Jul 20251 year ago | A small public preview; ARC Prize published 30-day learnings on 2025-08-19. |
Sources: each benchmark's paper (arXiv) and its official site or leaderboard. Scores are the boards captured here every day, exactly as published.