OpenAI no longer has a standalone robotics team — and according to the company, that is not a retreat. It is a bet that the same general-purpose AI models behind its chatbots have become good enough to run robots directly, with no specialist robot brain required.

A humanoid robot and a robotic arm in a futuristic AI lab beneath a glowing neural network brain, illustrating general AI models controlling robots

The change surfaced in an odd way. Fortune, which broke the story in an exclusive report on November 13, found OpenAI’s reserved booth at the Conference on Robot Learning (CoRL) in Austin standing empty, even though the company was still listed on the floor plan. OpenAI then confirmed the reorganization to Fortune.

Where the robotics team went

Under the new structure, OpenAI’s robotics researchers have moved into the company’s main research organization, led by Chief Research Officer Mark Chen. Its applied robotics and hardware staff have joined the Applied organization, led by President Greg Brockman.

An OpenAI spokesperson told Fortune the move reflects a change in approach, not a withdrawal from robotics. The company says general-purpose frontier models — including GPT-6 Astra — have shown they can be applied to robotics directly. That is the same agentic direction we covered when OpenAI launched ChatGPT Wallet, letting AI agents act on a user’s behalf:

The 74.8% number behind the bet

The evidence OpenAI is leaning on is striking. In a study by Nvidia researchers, an agent built on GPT-6 Astra scored 74.8% on RoboLab-120, a simulated benchmark of 120 tabletop robot tasks — well ahead of the 51.0% posted by the best specialized robot policy on that leaderboard.

There is a big catch, though. Those results are from simulation, not physical robots, and Astra’s direct calls averaged 18.56 seconds each while the simulation paused to wait for it. A real factory line does not pause. Specialist models are far faster, which is why many researchers see a split future: a general model plans, and a fast specialist model moves the arm.

This is also not OpenAI’s first robotics reset. It wound down an earlier team in 2020, rebuilt one in 2024, and partnered with humanoid maker Figure before that partnership ended in February 2025.

Why it matters

If OpenAI is right, robotics stops being a separate AI discipline and becomes another job for one big general model — a huge shift for every company building humanoids and factory robots. CEO Sam Altman is still targeting a first public humanoid robot demonstration in 2027, now with the work spread across the wider company rather than one dedicated team.

The timing is spicy, too: the reorganization lands in the same week the industry is fighting over the future of frontier AI itself, as we reported in our coverage of the lawsuit seeking to freeze stronger AI models.

For now, the scoreboard is simulation-only. The real test is whether a general model can move from winning benchmarks to moving real metal — quickly, cheaply, and safely.

Source: Fortune (exclusive, November 13, 2026), as reported by multiple outlets.

Mohsen Daemi