OpenAI had a robotics team in its early days but wound it down in 2020, with cofounder Wojciech Zaremba later citing limited robot data and a more efficient path toward artificial general intelligence (AGI)—the human-like AI that has been OpenAI’s goal since its founding—through other methods. In particular, OpenAI became increasingly devoted to training ultra-large language models on vast datasets of human-written text. This is the approach that eventually yielded the company’s breakthrough product ChatGPT in late 2022.

OpenAI rebuilt a stand alone robotics team in 2024, announcing a partnership with the startup Figure to develop AI models for humanoid robots. That partnership ended in February 2025, when Figure said it would develop its own robot AI. OpenAI has said it is currently developing its own humanoid robot, with CEO Sam Altman targeting a first public demonstration in 2027.

The company spokesperson said that despite the reorganization, OpenAI remains committed to bringing general-purpose robots into the world.

The reorganization reflects a bet that is gaining traction across the AI industry: that general-purpose models built for language, code, and vision can also be used for “embodied AI”—where AI models control devices that can act in the physical world, such as robots, driverless cars, or drones. This would obviate the need to have separate robotics teams building specialized AI models just to be the digital “brains” for robots.

Researchers have seen surprising performance when connecting GPT-6 Astra to robots, directly using the general-purpose AI model to perform some robotic tasks.

Guanya Shi, an assistant professor at Carnegie Mellon University’s Robotics Institute who had previously heard about OpenAI’s reorganization, told Fortune that he would interpret the move as a signal that OpenAI sees robotics as “no longer a special problem.” Instead, he said, they likely see robotics as a challenge that can be addressed as a downstream application of general intelligence—similar to the way the same underlying frontier AI models now power specialized coding tools, or tools for legal or financial work.

OpenAI seemed to confirm this view in its response to Fortune. It said that at the start of the year, the company’s researchers assumed the best way to make progress on robotics was to build on its work on AI models such as Sora, a video generation model that, in order to create realistic videos, had to have some implicit understanding of how the physical world works.

But since then, the company spokesperson said, the company’s general-purpose frontier models, including GPT-6 Astra, have shown that can be directly applied to robotics. This discovery is what motivated OpenAI to disband the stand alone robotics team and move the affiliated researchers onto teams working on more general AI models.

The reorganization also comes at a time when OpenAI is under pressure to rein in its costs and show that it is more focused on a few, core products. Last year, Altman told the company that it needed to abandon “side quests,” and the company has jettisoned some efforts—including Sora—in order to reallocate computing resources to more core and potentially-profitable endeavors. The company is under pressure to stanch its heavy operating losses—which reportedly totaled $10.9 billion in the most recent quarter—as it seeks to close a new funding round and prepares for a possible IPO in early 2027.

This line of thinking—that general AI models can be applied directly to controlling robots without the need to create specialized models with “spatial intelligence”—would seem to challenge the assumptions underpinning many of today’s robotics startups, which have raised billions on the premise that robots need brains of their own. 

But Russ Tedrake, an MIT professor and cofounder and CEO of Walden Robotics, said robotics companies still have an advantage the AI labs don’t: Their robots are already gaining work experience in real places, such as on factory floors. Such “deployment” experience—and the data these robots gather—helps robots adapt to specific workplaces and tasks. Better foundational models make that site-specific data more valuable because robots can learn from it faster, Tedrake said.

Recent experiments have tested how much work a general-purpose model can do without specialized training on robotic data. Researchers gave Astra camera images, robot information and a control interface that allowed the model to command the robot to take certain actions. In an Nvidia study, an Astra-powered agent achieved a 74.8% success rate on a simulated manipulation benchmark—where the model commands a simulated robot in a digital simulation, as opposed to a real robot—compared with 51.0% reported by the best specialized robotic models on this leaderboard.

Shi said advances in general models’ spatial and reasoning capabilities already let them handle tasks that previously required dedicated training. They can cope when a bottle is in a different position or the lighting changes, and reason about the sequence of a task or which step went wrong, he said.

Meanwhile, large language models are powerful but slow and need to be hosted in the cloud, meaning the robot or device needs a constant network connection. Shi still sees a need for smaller, specialized models that can run on a robot’s onboard computer without needing a constant internet connection.

“There may be different starting points for how to grow intelligence,” Pete Florence, cofounder of Generalist, a startup training general-purpose AI models for robots, told Fortune. He said it was unclear that any existing model would be the one that dominates robotics in the long-run.

For its part, OpenAI said it was reassigning its robotics researchers to other teams because it now believes this is the best way to accelerate progress towards building general-purpose robots.

Back at booth #208 at CoRL, a few former OpenAI robotics team members still drew a crowd, despite having no display or demo. After all, one of the conference’s hottest questions was how much robotics expertise would remain valuable as general-purpose models grew more capable—and the experience of those former OpenAI robotics researchers, now dispatched to other teams, offered one answer.