What it measures
- Abstract reasoning
- Few-shot generalization
- Efficiency-sensitive problem solving
ARC-AGI-2
A hard abstract-reasoning benchmark from ARC Prize focused on novel task generalization.
How to read the score
score; higher is better.
Sourced Ranking
| Rank | Model | Provider | Score | Source | Source Type | Verified |
|---|---|---|---|---|---|---|
| 1 | Gemini 3.1 Pro | 77.1 | Gemini 3.1 Pro announcement Accessed 2026-06-12 | official | Yes |
Limitations
- Scores should be read with the stated ARC Prize evaluation and efficiency rules.
- arXiv or technical papers are static; use the ARC Prize leaderboard for scores.
- Official leaderboard is the current score source. Track eval-set and cost/efficiency notes.
leaderboard
ARC Prize official leaderboard
arXiv or technical papers are static; use the ARC Prize leaderboard for scores. Official leaderboard is the current score source. Track eval-set and cost/efficiency notes.
official
ARC-AGI-2 paper or specification
Use for benchmark definition, not necessarily for latest model scores.
third-party
ARC-AGI-2 dataset or code
Use for dataset, code, or implementation details; score freshness depends on the benchmark.