LLMs aren’t just disrupting professions and jobs — they’re also disrupting other areas of AI research.
OpenAI has broken up its stand-alone robotics team and moved its researchers onto other teams, in a reorganization that reflects a growing belief in the industry: the same frontier models that write code and answer questions may be able to drive robots too, with no specialized “robot brain” required.
The company confirmed the change to Fortune. Its robotics research team now sits within the main research organization under Chief Research Officer Mark Chen, while the applied robotics and hardware teams have moved to the Applied organization, which reports to President Greg Brockman. An OpenAI spokesperson said robotics remains central to the company’s long-term plans and described the move as a sign of commitment rather than retreat.
Why OpenAI says it made the move
According to the company, researchers began the year assuming the best route to better robotics was to build on work like Sora, whose video generation required some implicit grasp of how the physical world behaves. Since then, OpenAI says, its general-purpose frontier models, including GPT-6 Astra, have shown they can be applied to robotics directly. That finding is what prompted the decision to fold the team into broader model-building efforts.
Early results back the claim. In a study run by Nvidia, an Astra-powered agent was given camera images, information about the robot and a control interface, and it succeeded on 74.8% of tasks on a simulated manipulation benchmark. The best specialized robotics models on the same leaderboard managed 51.0%. It’s a simulation rather than a physical robot, but the gap is large. Astra has also been showing its physical-world chops elsewhere: it recently became the first model to drive a real car through a cone course, and OpenAI has pitched it as a model built for computer use and real workplace tasks.
Guanya Shi, an assistant professor at Carnegie Mellon’s Robotics Institute, told Fortune he reads the move as a sign that OpenAI sees robotics as “no longer a special problem” and instead as a downstream application of general intelligence, much like the way one underlying model now powers coding tools and legal or financial products. Shi says general models can already cope with things that once demanded dedicated training, such as a bottle sitting in a new spot, a change in lighting, or working out which step of a task went wrong.
A long and winding robotics history
OpenAI has been here before. It had a robotics group in its early days but wound it down in 2020, with cofounder Wojciech Zaremba later pointing to scarce robot data and a more efficient path to AGI through large language models. That bet produced ChatGPT.
The company rebuilt a dedicated team in 2024 and partnered with Figure on AI for humanoid robots. Figure chose to build its AI in-house and the partnership ended in February 2025. OpenAI says it is now developing its own humanoid, and Sam Altman is targeting a first public demonstration in 2027, a year he has previously described as when AI moves from the intellectual realm into the physical world. He has also floated the idea that the top ChatGPT tier might one day come with a humanoid robot.
Cost pressure in the background
The reorganization also lands while OpenAI is trying to tighten its focus. Altman told staff last year that the company needed to drop “side quests,” and OpenAI has since cut projects, including Sora, to free up compute for core products. That clean-up has also cost it some senior leaders. Fortune reports the company is under pressure to limit heavy operating losses, which reportedly reached $10.9 billion in the latest quarter, as it works to close a new funding round and prepares for a possible IPO in early 2027, even though Altman has said it won’t list in 2026.
What it means for robotics startups
If one general model can steer a robot, the pitch behind many robotics startups, which have raised billions on the idea that robots need purpose-built brains, looks shakier. Not everyone agrees. Russ Tedrake, an MIT professor and the cofounder and CEO of Walden Robotics, argues that robotics companies hold an edge the AI labs lack: their machines are already working in real places like factory floors, and that deployment experience and data help robots adapt to specific tasks. In his view, better foundation models make that site-specific data even more valuable, since robots can learn from it faster.
Shi also sees a continuing need for smaller, specialized models. Large language models are slow and typically hosted in the cloud, which means a robot depends on a constant network connection, whereas a compact model running on the robot’s own computer doesn’t. Google has been pursuing both ends of this, with a robotics model built to run on-device and a recent release that used Gemini 4 Argon to generate simulated training data.
For OpenAI’s part, it says the reassignment is simply the fastest way to get to general-purpose robots. The researchers who once made up its robotics team will now be testing that claim from inside the teams building the models themselves.