Unverified claim: GPT-6 Astra scores 91.8% on SpatialBench, beating 80% human baseline

paigeinsf · x · 2026-09-08

A viral post claims "GPT-6 Astra is AGI," citing a 91.8% score on the spatial reasoning benchmark SpatialBench vs an 80% human baseline. A replier doubts the AGI framing but agrees spatial reasoning is a big step for robotics. Unverified, from a personal account with no official confirmation.

Original post →

More from Models

Models channel →