X.PIN· @thexpin · X·· 5 小时前AI 评分67
AI 导读
清华大学与 UC Berkeley 研究团队开源 uni-robot-agent 框架，让 Qwen3.8-Flash-Next 等开源模型通过统一工具接口在零样本机器人操控任务中超越 GPT-6 Astra。实测显示其单任务成本约为 1 美元，仅为 OpenAI 的 6.6%。
正文 · 原文
GPT-6 Astra just became the ultimate target for China's Embodied AI scene.
Researchers from Tsinghua and UC Berkeley (incl. Pieter Abbeel) open-sourced uni-robot-agent, allowing local Qwen3.8-Flash-Next to beat GPT-6 Astra in zero-shot robot control, at just ~$1 per task (6.6% of OpenAI's cost).
来源：X.PIN · x.com