GPT-6-Astra as Robot Policy Tops RoboMME, Beating Human Baseline

Harvard researchers show a frozen GPT-6-Astra with full visual history in context can directly serve as a memory-dependent robot manipulation policy, topping the RoboMME benchmark at 92.1% success, above the 90.5% human baseline.

2026-10-17 ~ 2026-10-17 · 2 related posts

• Embodied Astra tops RoboMME at 92.1% — beating humans by just keeping full visual history in context — hankyang94 · 2026-10-17
• Frozen GPT-6-Astra used directly as robot policy hits 92.1% on RoboMME, beating human baseline — hankyang94 · 2026-10-17