GPT-6 Astra Scores 95% on Robot Control, 2.3x Cheaper Than Fable 5.1

willcb · x · 2026-09-06

A quoted thread reports GPT-6 Astra scoring 95% on a robot control task, up from Fable 5.1's 40%, while using 6.2x fewer output tokens at 2.3x lower cost. A follow-up question asks whether GPT-6 could simply API-call the MolmoAct2 checkpoint to execute commands like "put block into bowl," touching on whether general-purpose models can replace specialized robot policies.

Related event: GPT-6 Astra scores 95% on robot control task at 43% of prior cost(5 posts)→

Original post →

More from Embodied

Embodied channel →