gpt-6VIDEO

BiGym 2.0 家用机器人基准：GPT-6 Astra 单次演示达 65%

新基准测试显示 GPT-6 Astra 用 1 次演示达 65%，但用 60 次演示训练的 π0.5 仍以 75% 领先，学习型策略尚未被取代。

Jesse Zhang

@Nag1ovo24 天前收录原推发布于 2026.10.09

Jesse Zhang @Nag1ovo原推发布于 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/2108230199187808367采集于 2026.10.09 10:38