OpenAI Disbands Stand-Alone Robotics Team, Shifts Robot Control to General-Purpose Models
OpenAI disbanded its stand-alone robotics team.
Research reports to Mark Chen; applied hardware to Greg Brockman. It says general models can now control robots.
Nvdium Tech | November 14, 2026
OpenAI folds its robotics team into the main research and applied groups
OpenAI has disbanded its stand-alone robotics team and reassigned the people on it to other parts of the company. The company confirmed the reorganization, saying robotics remains a core part of its long-term plans and that the change is meant to speed progress, not walk away from the field. Robotics research now sits in the main research organization, reporting to Chief Research Officer Mark Chen. Applied robotics and hardware have moved into the Applied organization, which reports to President Greg Brockman. At this week's Conference on Robot Learning (CoRL) in Austin, the booth OpenAI had already reserved stood empty.
OpenAI had a robotics effort in its early years, wound it down in 2020, and rebuilt a stand-alone team in 2024, including a partnership with humanoid-robot startup Figure. That partnership ended in February 2025, when Figure said it would develop its own robot AI. OpenAI has kept work on its own humanoid robot, with CEO Sam Altman aiming for a first public demonstration in 2027. The company said that at the start of this year, its researchers assumed the best path was to build on work such as Sora, a video model that has to understand something about how the physical world works. Since then, general-purpose frontier models including GPT-6 Astra have shown they can be applied directly to robot control, which is what prompted the company to drop the stand-alone team.
In NVIDIA's RoboLab-120 simulation benchmark, an agentic policy that uses GPT-6 Astra in a persistent session reached a 74.8% success rate, above the 51.0% posted by Lakeside, the strongest specialized policy on that leaderboard. The results are from simulation, and a direct call to the model still takes about 18.56 seconds on average, too slow for real-time control on a physical robot. OpenAI said it is moving robotics researchers onto the teams that build general models because it now sees that as the faster route to general-purpose robots.