Study suggests GPT-6 Astra's robot skills come from specialized training data, not emergent general intelligence
A technical analysis speculates that GPT-6 Astra's robot-manipulation abilities may stem from specialized robot training data rather than emergent general intelligence. The study points to demos with NVIDIA and Unitree, and notes that OpenAI has not disclosed whether pre-training included robot-specific data. Similar training is suggested for GPT-6.1 Sol and Gemini Argon.
Coverage timeline
机器之心机器之心
2026 年 9 月 3 日，GPT-6 Astra 正式发布。 除了推理、编程和 Agent 能力，这款新模型在机器人操作领域展现出的潜力，也引起了不少关注。 在很多人看来，一个通用多模态大模型，似乎已经能够理解机器人所处的环境，并根据任务要求生成相应动作。 发布后，NVIDIA 将 GPT-6 Astra 引入机器人仿真与设计工作流。在一项演示中，开发者使用 Astra、PTC Onshape 和 NVIDIA Isaac Sim，建立汽车悬架的 3D 模型，调整扳手的形状与位置，使其能够触及悬架上的螺栓，最终在仿真环境中演示机器人使用扳手拆卸部件的过程。 来源：https://x.com/nvidiaomniverse/status/2118450742176698700 另一项演示借助 Unitree G1 人形机器人，Astra 可以在一个从未见过的房间里自主行动，记住不同物体的位置，跨越整个房间完成整理任务。之后，即使人类只给出比较模糊的自然语言指令，它也能根据此前建立的空间记忆，找到并取回指定物品。 来源 ：https://www.reddit.com/r/singularity/comments/1wqwcqe/gpt6_astra_can_now_control_a_humanoid_robot_in_a/ 类似的案例还有不少，这里就不一一列举了。 GPT-6 Astra 一出，机器人圈就热闹了起来。有人直接喊出：VLA 已死，Robot RSI 时代来了！通用大模型要接管机器人了！ 但一个更关键的问题来了：GPT-6 Astra 展现出的机器人能力，究竟是通用智能发展到一定阶段后自然涌现的结果，还是模型在训练过程中已经积累了大量机器人领域的专门知识？ OpenAI 官方博客并未明确披露模型训练数据的具体构成，也没有说明预训练阶段是否使用了专门的机器人数据。 有研究发现，GPT-6 Astra 的机器人能力并非涌现出来的 最近，一项研究给出了一个值得关注的线索。 来源：https://x.com/SwannAiden/status/2118741112227774702 该研究指出， GPT-6 Astra 的机器人能力并非涌现出来的，OpenAI 在训练过程中使用了专门的机器人数据。GPT-6.1 Sol 和 Gemini Argon 也接受了类
Related stories
SenseTime-led study: GPT-6-Astra scores near-perfect on benchmarks but fails real-robot tasks
A research report by SenseTime, Daxia Robotics, Ropedia, NTU Singapore, and the University of Tokyo analyzes GPT-6-Astra's embodied intelligence across 12 simulation benchmarks and 8 real-robot platforms. The model achieves near-perfect scores on traditional benchmarks (e.g., 100% on LIBERO-Long++, 99.3% on LIBERO-Pro) but fails on real-robot tasks, such as taking over 4 hours and $55 to place a pen in a holder, and failing 11 insertion attempts. The report concludes GPT-6-Astra has largely crossed the embodied intelligence threshold but remains far from practical deployment.
1 reports · 1 sources
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.
1 reports · 1 sources
MIT: GPT-6 Astra edits Qwen weights, boosting MATH-500 from 7.9% to 48.1%
MIT CSAIL researchers demonstrated that GPT-6 Astra can directly read and modify the weights of Qwen3.5-0.8B, improving its MATH-500 score from 7.9% to 48.1% by altering only 0.016% of parameters, without training or gradient descent. The team named this capability Weight-Space Programming. However, spaCy author Matthew Honnibal publicly challenged the paper, arguing that the experimental conditions across the three claimed abilities differ significantly and that the presentation is misleading.
6 reports · 5 sources2949 points · 892 comments
OpenAI GPT-6.2 Astra Spotted in Codex Usage Stats, Official Listing Absent
User screenshots and reports indicate OpenAI may be conducting a wide gray-area rollout of a model internally labeled GPT-6.2 Astra, appearing in Codex usage statistics while the official model catalog still lacks it. The model may skip the reportedly canceled GPT-6.1 Astra, with community speculation suggesting fixes to prior safety issues and a version bump. Reports claim performance gains over GPT-6 Astra and lower costs, but official confirmation is pending.
News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe