ARC-AGI 3 Leaderboard 2026 — Results Across 4 Real AI Models

ARC-AGI 3 leaderboard

ARC-AGI 3

4 models tested · Updated 2026-03-24 · Verified sources only

      Gemini 3.1 Pro Preview leads at 0.37%
    

Google · ARC Prize/arcprize.org · 2026-03-24

Highest score on new interactive reasoning benchmark. Humans score 100%. ARC-AGI-3 uses turn-based game environments with no instructions.

0.37%

OpenAI · ARC Prize/arcprize.org · 2026-03-24

Second-highest on ARC-AGI-3. All frontier models score below 1% on this new interactive reasoning benchmark.

0.26%

Anthropic · ARC Prize/arcprize.org · 2026-03-24

Third on ARC-AGI-3. The gap between humans (100%) and AI (best 0.37%) on interactive reasoning remains enormous.

0.25%

xAI · ARC Prize Foundation · 2026-03-25

Only frontier model to score exactly 0% on ARC-AGI 3 at launch. Exceeded action cutoff on every level. Humans score 100%.

0.0%