Ecosystem

OpenAI disbands its robotics team because GPT-6 Astra can run robots directly

2 min read AI-generated

At the CoRL robotics conference in Austin, OpenAI's booth 208 sat empty. The researchers now report to Mark Chen and Greg Brockman.

Featured image for "OpenAI disbands its robotics team because GPT-6 Astra can run robots directly"

OpenAI has broken up its stand-alone robotics team. Fortune reported it as an exclusive, and a spokesperson confirmed it when asked. Robotics remains a core part of the company’s long-term vision, the spokesperson said, and the move signals commitment rather than retreat.

Where the people go

The robotics research group moves into the main research team under Chief Research Officer Mark Chen. The applied robotics and hardware teams join the Applied organization, which reports to President Greg Brockman. The decision seems to have been recent. At CoRL, a top academic robotics conference running in Austin this week, OpenAI was still listed at booth 208 in the floor plan. When Fortune stopped by, the spot was empty, no logo in sight.

The reason is GPT-6 Astra

At the start of the year, according to the spokesperson, OpenAI’s researchers assumed robotics progress would come through models like Sora, a video model that had to pick up an implicit sense of physics. Since then, the general-purpose frontier models, GPT-6 Astra among them, have turned out to apply to robots directly. Fortune cites an Nvidia study in which an Astra-powered agent hit a 74.8% success rate on a simulated manipulation benchmark, against 51.0% for the best specialized robotics models on that leaderboard. Astra got camera images, robot information and a control interface.

Guanya Shi of Carnegie Mellon’s Robotics Institute reads the move as OpenAI treating robotics as “no longer a special problem.” Robotics would then be a downstream application of general intelligence, the way coding tools and legal or finance tools are today.

Third attempt in six years

OpenAI had a robotics team once before and shut it down in 2020 for lack of data. It rebuilt one in 2024, with a Figure partnership that ended in February 2025. Sam Altman wants to show off an in-house humanoid in 2027. Fortune also notes the cost pressure: OpenAI reportedly lost $10.9 billion in its most recent quarter, and Altman had told staff to drop “side quests.”

The math gets tighter for robotics startups

This bet hits a whole industry. Plenty of robotics startups raised billions on the premise that robots need brains of their own. MIT professor Russ Tedrake, co-founder and CEO of Walden Robotics, pushes back: companies with robots already deployed are collecting experience in real workplaces, and better base models make that experience more valuable. I find that the stronger argument. A simulated benchmark is a long way from a factory floor, and the factory data is something OpenAI doesn’t have.

Sources

OpenAIRoboticsGPT-6 AstraEmbodied AI