Benchmark

ARC-AGI-3

Interactive multi-step format, so the agent harness is part of the measurement. Scores are low and spread wide, which makes small absolute gaps look…

Released
2026-01-15
Categories
reasoning
Openness
unknown
Source
model_reports
Reported scores
1

Reported scores

model_reports

ModelOrganizationReported valueReportedEvidence
Claude Opus 5Anthropic30.162026-07-24evidence

Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.