The hardest benchmark starts to move

Claim

GPT-5.5 scored 0.43% on ARC-AGI-3's semi-private set, more than double Opus 4.7's 0.18%, making abstract fluid reasoning look less like a wall than a ramp.

Reported metric: 0.43% vs 0.18%

Reporting date: 2026-05-03 (not necessarily the event date)

Sources

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

• Welcome to May 3, 2026 (theinnermostloop.substack.com)