A model reaches the human baseline on a benchmark built to be impossible

Claim

On ZeroBench, the impossible visual benchmark, GPT-5.6 Sol became the first model to reach the 30% human baseline, edging Claude Opus 5 at 26% and Fable 5 at 24%.

Reported metric: 30% human baseline

Reporting date: 2026-08-12 (not necessarily the event date)

Sources

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

• Welcome to August 12, 2026 (theinnermostloop.substack.com)