On November 13, according to foreign media reports, OpenAI has disbanded its independent robotics team, with its former members folded into the regular model training system. Internal assessment concluded that robotics is merely one type of vertical-domain task for general-purpose models, and the intelligence capabilities it requires should be addressed uniformly within the general foundation model framework. As of press time, OpenAI has not officially confirmed or disclosed relevant details.
This is not the first time OpenAI has adjusted its robotics path. In 2019, its robotic hand Rubik's Cube-solving project, Dactyl, attracted widespread attention; in October 2020, the company disbanded its robotics team internally, and in 2021 co-founder Wojciech Zaremba publicly explained that physical interaction data was difficult to scale, so resources were redirected to large models. In the following years, OpenAI's approach was summarized as "only build the brain, not the body."
From late May to June 2026, Sam Altman publicly announced hiring for OpenAI Robotics, a team that evolved from Worldsim; in September, the company confirmed it was developing its own humanoid robot; in October, hardware roles continued to expand, and the logic shifted to "building its own body and mastering the data closed loop." If this rumor is true, it means OpenAI will no longer retain a standalone robotics organizational unit, but hardware and hiring have not come to a complete halt.
From a strategic perspective, the robotics direction appears to still be retained, but its positioning has shifted from an "independent legion" to "an application scenario for foundation models," reflecting that the general model route is extending into the field of embodied intelligence. At present, this news has yet to be officially confirmed.