gpt-6VIDEO
BiGym 2.0 household robot benchmark: GPT-6 Astra reaches 65% from one demo
A new benchmark shows GPT-6 Astra reaches 65% with one demo, but π0.5 trained on 60 demos still leads at 75%, so learned policies have not been replaced yet.
Jesse Zhang
@Nag1ovoArchived 24 days agoOriginally posted 2026.10.09
Jesse Zhang @Nag1ovoOriginally posted 2026.10.09
Coding agents can now write a humanoid's controller from one demo. So are learned policies dead? We built BiGym 2.0, a household loco-manipulation benchmark, to find out. Short answer: not yet. GPT-6 Astra hits 65% with 1 demo, but π0.5 trained on 60 demos still leads at 75%, especially on precise object rearrangement. 🧵
x.com/nag1ovo/status/2108230199187808367Archived 2026.10.09 10:38