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.

GPT-6 Astra by OpenAI holds the best ARC-AGI-2 result among tracked models, at 95% (released Sep 3 2026). Higher scores are better on this benchmark.

1. GPT-6 Astra (OpenAI) — 95%; 2. GPT-5.6 Sol (OpenAI) — 92.5%; 3. Claude Opus 5.5 (Anthropic) — 91.7%; 4. Claude Opus 5 (Anthropic) — 90.4%; 5. Claude Fable 5.1 (Anthropic) — 90%.

34 tracked models have a published ARC-AGI-2 score. Scores are collected from published lab reports and benchmark sources, with recorded sources credited alongside the results. We do not re-run the evaluations.

DeepSeek-V4-Flash-0731 by DeepSeek is the highest-ranked model with downloadable weights on ARC-AGI-2, scoring 61.4% at rank 19 overall.