GPT-6 Astra Reportedly Tops ZeroBench, Beating Human Baselines

GPT-6 Astra (max) reportedly scores 51% pass@5 on the hard visual benchmark ZeroBench, far exceeding the prior SOTA of about 30% and surpassing human baselines on all three metrics (unverified).

2026-09-23 ~ 2026-09-24 · 2 related posts

• GPT-6 Astra tops ZeroBench, surpassing human baseline across all three metrics — Waiting4AniHaremFDVR · 2026-09-23
• GPT-6 Astra (max) reportedly hits 51% pass@5 on ZeroBench, far above 30% SOTA — scaling01 · 2026-09-24