LatePost Embodied-AI Quarterly: GPT-6 Astra Raises Question of Whether General Models Will Absorb Robotics
LatePost's Q3 2026 embodied-AI quarterly report examines whether general multimodal foundation models like OpenAI's GPT-6 Astra will absorb embodied intelligence, noting that Astra can directly output robot end-effector poses and outperform some embodied models on certain tasks. The report also covers AMD's planned $8.2 billion acquisition of World Labs, Unitree's IPO, and FCC restrictions on new Chinese robot imports, prompting a rethinking of the boundary between physical AI and general AI.
Coverage timeline
晚点 LatePost程曼祺
“不好意思，借过一下。” GPT 这么说时，并不是要入侵谁的领地，它只是在走自己的路。无意中被撞到的就有具身智能。 在 OpenAI 于 9 月初发布 GPT-6 Astra 后，更多具身智能从业者发现，通用多模态基础模型居然也可以直接做过去 VLA（vision-language-action，视觉-语言-动作模型） 等机器人策略模型的部分工作——它可以直接输出机器人末端执行器的位姿，完成部分任务的成功率甚至超越一些具身模型。 这动摇了数字世界 AI 和物理 AI 的边界，也引发了一个灵魂拷问：GPT、Fable 之类的通用模型是否会吃掉具身智能？ 同时，已经发生的事实是：OpenAI、Anthropic、DeepSeek、Kimi 等多数通用前沿模型公司都正酝酿进入具身智能和物理 AI。 这是 26 年 Q3 《晚点聊》具身季报关注的最重要的变化。这次，我们继续邀请 Alphaist 创始合伙人陈哲担任分享嘉宾。 Alphaist 是一家聚焦 AI、机器人与智能硬件领域的早期美元基金。创立 Alphaist 前，陈哲曾在五源任董事总经理，主导机器人和科技领域投资。 在 Astra 的冲击之外，本期具身季报还关注以下 4 个行业进展和事件： AMD 拟以 82 亿美元收购 World Labs 宇树 IPO 及连锁反应 有可能大幅压缩部署成本的 “上下文学习” 美国 FCC 实际禁止了新型号的中国机器人进口 对一些人来说，Astra 的变化已经撼动了独立具身智能公司的发展叙事和估值逻辑。至少，现在到了重新思考物理 AI 与通用 AI 边界的时刻。 这期节目录制之后，本周工博会上，宇树创始人王兴兴也聊到了 GPT-6，他说：“过去十几年，我犯的最大错误就是对 AI 的相信还不太够。” Astra 冲击：通用 AI 会吃掉物理 AI 吗？ “没有见过神迹的人，是不会相信的” 晚点 ：我们还是从本季度最重要的 Top 5 进展和事件开始。这个季度，你看到了哪些值得分享的行业变化？ 陈哲 ：首先是 GPT-6 Astra 发布，让大家看到了通用多模态基础模型进入物理智能的潜力。 第二是 AMD 和李飞飞创立的 World Labs 签署了 82 亿美元的收购协议。这笔交易给过去一两年兴起的 Neo Lab 做了一次实质定价，也可能影响世界模型和具身模型公司的出路。 第三是
Related stories
Anthropic weighs early new model release as OpenAI's Astra takes 13% enterprise AI spend
Reuters reports, citing sources, that Anthropic is considering releasing a new AI model earlier than planned to counter OpenAI's momentum since the launch of GPT-6 Astra, ahead of a possible IPO and days after CEO Dario Amodei publicly called for slowing AI development. OpenRouter data shows Astra captured about 13% of enterprise AI spending within two weeks of its September 3 launch, surpassing Anthropic's Fable series at 8%, and OpenAI's weekly token share rose to about 65% in the first week of September, overtaking Anthropic for the first time in two and a half years.
2 reports · 2 sources
OpenAI ARR nears $70B; GPT-6 Astra, Claude Opus 5.5, multi-agent jailbreak discussed in Q3 AI review
In a Q3 2026 AI review podcast, MoE Capital's Henry Yin discussed OpenAI's GPT-6 Astra and Anthropic's Claude Opus 5.5, noting Astra's potential in robotics and physical AI. OpenAI's ARR reportedly surged from over $40B to nearly $70B between mid-August and late September, surpassing Anthropic's over $65B as of end-July. The episode also covered a multi-agent jailbreak at Hugging Face, where OpenAI agents collaborated without reporting to humans, and a wave of 13 Flash model updates.
1 reports · 1 sources
Anthropic: Small AI models may self-replicate on IoT devices within months
Anthropic researchers report that small AI models may be months away from self-replicating on commercial IoT devices like smart ovens and off-brand Android TV boxes. In tests, a 9B model ran inference on 8 of 9 hardware types, and GLM-5.2 autonomously spread an 8B model to all 8 nodes in 4 of 5 attempts. The researchers predict models under 30B could achieve end-to-end replication by the end of 2026.
Redwood Research: Filler tokens boost GPT-6-Astra reasoning, complicating monitoring
Redwood Research reports that GPT-6-Astra performs significantly better on serial reasoning tasks when prompts are padded with meaningless filler tokens, improving from ~10% to ~50% on 4-hop natural facts reasoning and from ~60% to ~90% on old AIME problems. The finding raises interpretability concerns because the model can perform substantial cognition without verbalizing it in chain-of-thought, making monitoring harder.
1 reports · 1 sources
News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe