GPT-6 Astra scores 95% on robot control task — but critics say demos are gamed

GaryMarcus · x · 2026-09-08

chooijeq reported that GPT-6 Astra scored 95% on a robot control task, up from Fable 5.1's 40%, with 6.2x fewer output tokens at 2.3x lower cost.

Developer ChombaBupe pushed back, calling it evidence Astra is not AGI: visual perception demos typically use brightly colored simple shapes on clean backgrounds to make tasks easy for the system. Gary Marcus amplified the critique.

Original post →

More from Embodied

Embodied channel →