OpenAI Disbands Its Robotics Team, Betting GPT-6 Astra Can Run Robots Directly

At CoRL, the big robot-learning conference in Austin this week, the floor plan still showed an OpenAI booth at #208. When a Fortune reporter walked over, it was empty, with no OpenAI logo in sight. The reason: OpenAI had just disbanded its stand-alone robotics team.

Fortune broke the story on Nov. 13, and OpenAI confirmed the reorganization. Robotics researchers now sit in the main research team under Chief Research Officer Mark Chen. The applied robotics and hardware teams move into the larger Applied organization run by President Greg Brockman. An OpenAI spokesperson said the company isn't leaving robotics. It called robotics a core part of its long-term vision and described the change as a way to move faster. OpenAI has said it is building its own humanoid robot, and CEO Sam Altman is aiming for a first public demo in 2027.

This isn't the first time OpenAI has broken up a robotics group. It worked on robots in its early years and wound that effort down in 2020. Cofounder Wojciech Zaremba later pointed to limited robot data, and to large language models as a more efficient path to artificial general intelligence. OpenAI rebuilt a stand-alone robotics team in 2024 and partnered with humanoid startup Figure. That partnership ended in February 2025.

This time the reasoning is different. OpenAI told Fortune that at the start of the year its researchers thought the best route to robots ran through video models like Sora, which have to understand how the physical world behaves. Since then, the company says, general frontier models including GPT-6 Astra have proven they can be applied directly to some robotics tasks. So the robotics researchers are joining the teams that build those general models.

What does applying a model directly look like? Researchers give Astra camera images, information about the robot and a control interface, then let it command the robot itself. Fortune cites an Nvidia study in which an Astra-powered agent scored a 74.8% success rate on a simulated manipulation benchmark, compared with 51.0% for the best specialized robot models on that leaderboard. That test ran in a computer simulation, not on a physical robot.

Here's one way to picture it. The industry has assumed robots need a dedicated brain, the way a company keeps a separate translation department. OpenAI is betting its generalist employees now speak enough languages that the department can be dissolved and its people spread across the business. Guanya Shi, an assistant professor at Carnegie Mellon's Robotics Institute, told Fortune he reads the move as OpenAI seeing robotics as "no longer a special problem," but a downstream use of general intelligence, the same way one frontier model now powers coding, legal and finance tools.

That's the new part. For the past few years, the standard approach has been to train dedicated vision-language-action models as robot brains, and robotics startups have raised billions on that premise. OpenAI is now reorganizing around the opposite bet: skip the separate robot brain and put the general model to work.

The pushback is concrete. Shi notes that large models are slow and live in the cloud, so a robot needs a constant connection. He still sees a need for smaller, specialized models that run on the robot itself. MIT professor Russ Tedrake, cofounder and CEO of Walden Robotics, says robot companies have something AI labs don't: robots already gaining work experience on factory floors. Better foundation models make that site-specific data more valuable, not less, he argues.

Fortune also notes the timing. OpenAI is under pressure to drop "side quests" and rein in costs, with operating losses reported at $10.9 billion in the most recent quarter, as it works on a new funding round and a possible IPO in early 2027.

None of this puts a robot in your kitchen next month. But if OpenAI is right, teaching a robot a new chore could someday look more like a chatbot software update than training a brand-new specialized model. Household robots might arrive sooner than many expected, and they might also depend more heavily on a handful of big AI labs.

Original source published: Fortune, Nov 14, 2026, 01:05 Beijing time / 12:05 PM EST (Nov 13)