OpenAI has disbanded its standalone robotics team, pivoting its physical AI strategy to rely on general-purpose models instead of building specialized architectures for humanoid robots.
The robotics researchers have been reassigned to OpenAI's core research division under Chief Research Officer Mark Chen, while the applied robotics and hardware teams have joined the broader Applied organization led by President Greg Brockman. The transition is so recent that the company's registered booth at the CoRL robotics conference in Austin this week sat empty, devoid of branding or staff. This restructuring aligns with OpenAI's broader effort to cut costs and focus on core, profitable products ahead of a potential initial public offering targeted for early 2027.
The strategic shift is driven by the discovery that general-purpose frontier models, such as GPT-6 Astra, can be directly applied to physical tasks, bypassing the need for specialized embodied AI models. In simulated benchmarks, agents powered by these general models have achieved success rates of nearly 75%, significantly outperforming specialized robotic models. Proponents argue that the same underlying AI technology powering advanced language and coding tools can also handle the spatial reasoning required for physical robotics.
Despite the optimism surrounding general-purpose models, physical robotics still relies heavily on real-world deployment data gathered on factory floors, an area where dedicated robotics companies maintain an edge. Additionally, large cloud-based models face latency and connectivity constraints that require smaller, specialized on-board systems for seamless real-world operations. OpenAI remains committed to its humanoid robot project, with plans to showcase a public prototype by 2027.