GPT-6 Astra beats Claude Fable 5.1 on robot control: 35% vs 15% success

k7agar · x · 2026-09-18

Pantograph ran a controlled benchmark of GPT-6 Astra and Claude Fable 5.1 on Pandroid robots across eight manipulation tasks. Astra completed 35% of attempts vs 15% for Fable (p≈0.001), while human teleoperators completed every task — a rare head-to-head of frontier models on real robot control.

Original post →

More from Embodied

Embodied channel →