ARC-AGI v2
multimodal
+
+
+
+
About
ARC-AGI v2 is an enhanced version featuring newly curated tasks designed for more granular assessment of abstract reasoning and problem-solving capabilities. This improved benchmark provides wider scoring ranges, incorporates tasks less susceptible to brute-force solutions, and focuses on deeper human-like thinking in problem-solving. ARC-AGI v2 targets higher levels of fluid intelligence with empirically calibrated difficulty levels compared to human performance.
+
+
+
+
Evaluation Stats
Total Models4
Organizations4
Verified Results0
Self-Reported1
+
+
+
+
Benchmark Details
Max Score1
Language
en
+
+
+
+
Performance Overview
Score distribution and top performers
Score Distribution
4 models
Top Score
15.9%
Average Score
9.0%
High Performers (80%+)
0Top Organizations
#1xAI
1 model
15.9%
#2Anthropic
1 model
8.6%
#3OpenAI
1 model
6.5%
#4Google
1 model
4.9%
+
+
+
+
Leaderboard
4 models ranked by performance on ARC-AGI v2
| License | Links | ||||
|---|---|---|---|---|---|
| Jul 9, 2025 | Proprietary | 15.9% | |||
| May 22, 2025 | Proprietary | 8.6% | |||
| Apr 16, 2025 | Proprietary | 6.5% | |||
| May 20, 2025 | Proprietary | 4.9% |
