Abstract reasoning
ARC-AGI-2
Puzzle-style tests of abstract reasoning and pattern-finding — the kind of thing people find easy but AIs often struggle with. Higher is better.
Rankings
Higher is better184.6%
GPT-5.5
OpenAI · Apr 23 2026
283.3%
GPT-5.4-Pro
OpenAI · Mar 5 2026
377.1%
Gemini 3.1 Pro
Google · Feb 19 2026
475.8%
Claude Opus 4.7
Anthropic · Apr 16 2026
572.1%
Gemini 3.5 Flash
Google · May 19 2026
658.3%
Claude Sonnet 4.6
Anthropic · Feb 17 2026
753.3%
Grok 4.20 Beta
SpaceXAI · Feb 17 2026
852.9%
GPT-5.2
OpenAI · Dec 11 2025
942.5%
Muse Spark
Meta · Apr 8 2026
1033.6%
Gemini 3.0 Flash
Google · Dec 17 2025
1131.1%
Gemini 3.0 Pro
Google · Nov 18 2025