Benchmark

ARC-AGI-3

The ARC Prize Foundation's interactive reasoning benchmark of novel game environments, scored on how efficiently an agent acts relative to humans.

open-source

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to January 30, 2026 (theinnermostloop.substack.com)

ARC-AGI-3 is the ARC Prize Foundation’s third benchmark, an interactive set of game environments in which agents must acquire goals on the fly and are scored on action efficiency relative to humans (benchmark page). It enters this corpus with the announcement of that efficiency metric; at launch it returned frontier models to near zero with a best score of 0.37%, before an agent architecture lifted a model to a perfect score on all 183 levels in August.

Visit resource ↗

Development timeline

Newest reporting dates first. Reporting dates may differ from event dates.

2026-09-22: Communicating agents reduce the number needed for ARC tasks

2026-08-23: An agent architecture lifts a model from 30% to a perfect 100

2026-05-03: The hardest benchmark starts to move

2026-03-27: A new benchmark returns frontier models to near zero

Knowledge graphopen in browser →

Published by

ARC Prize Foundation

Referenced by

Cite or correct this record

Citation formatCitation text

ARC-AGI-3
Innermost Loop KB — secondary record
Reporting date: unknown
KB permalink: https://nicholsn.github.io/innermost-loop-kb/benchmarks/arc-agi-3
Attributed sources:
- Welcome to January 30, 2026 — https://theinnermostloop.substack.com/p/welcome-to-january-30-2026
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.