Should General-Purpose Large Models 'Dimensional Strike' Embodied Intelligence? The First AA-Robotics Ranking is Here
通用大模型要「降维打击」具身智能？首个AA-Robotics榜单来了
The AA-Robotics v0.1 evaluation for the first time places general-purpose large models in real robotic tasks, testing their ability to control robotic arms to perform tasks such as pouring water, writing, and building with blocks. Although GPT-6 Astra leads in overall scores, the model still faces challenges in stability, speed, and cost efficiency in physical interactions, highlighting the gap between language understanding and physical execution.
Source: Machine Heart — Robotics on WeChat · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
time
≈52 minutes
price
≈150
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
在新发布的 AA-Robotics v0.1 评测中，GPT、Claude、Gemini 等前沿模型需要控制真实机械臂，挑战倒水、写字、搭积木等八项任务，比一比谁的「动手能力」更强。
Open source S4
首批结果同样充满反差： GPT-6 Astra 排名第一 ，通用大模型已经能够借助编程 Agent 操控真实机械臂；但它们 完成一项任务平均需要约半小时 ，比人类基线 慢了将近两个数量级 。
Open source S8
五款模型中，除了 Gemini 4 Argon 平均耗时约 52 分钟，另外四款完成一项任务大约都要半小时。
Open source S23
对于未来要在家庭、仓库里连续工作的机器人来说，能完成一次，和能以合理成本每天完成上百次，是两回事。
Open source S24
一台售价约 150 美元的机械臂，在 Astra 的控制下，竟然能根据摄像头画面自主规划动作，一笔笔画出金门大桥。
Open source S11
Source：Machine Heart — Robotics on WeChat · mp.weixin.qq.com