# OpenAI disbands robotics team, says general models now run robots

> OpenAI dissolved its standalone robotics team, moving staff into research and applied units.

*Staff move into research and applied units as OpenAI bets its general models can control machines directly.*

By Bez · InsidersFeed
Canonical: https://insidersfeed.com/news/openai-disbands-robotics-team-says-general-models-now-run-robots

OpenAI has dissolved its standalone robotics team, a move that signals it will run robots through general models rather than a dedicated stack. [Fortune](https://fortune.com/2026/11/13/openai-disbands-robotics-team-as-its-general-models-start-running-robots/) reported the change on 13 November, calling it the company's second such move in six years.

Research staff now report to Chief Research Officer Mark Chen. Applied robotics and hardware staff report to President Greg Brockman. OpenAI wound down its first robotics effort in 2020, when cofounder Wojciech Zaremba cited "limited robot data". It rebuilt a standalone team in 2024, and its partnership with Figure ended in February 2025.

The case for the change rests on GPT-6 Astra. Fortune reports that it scored 74.8% success on a simulated manipulation benchmark, against 51.0% for specialised robotics models. On real hardware, it completed 7 of 100 bimanual desk tasks, while MolmoAct2 completed none. Each Astra robotics attempt cost $36.40 and took 18.56 seconds of latency. It is unconfirmed whether the benchmark results are public or internal.

Sam Altman is targeting a first public demo of OpenAI's own humanoid robot in 2027, according to Fortune. A company spokesperson said robotics "remains a core part of OpenAI's long-term vision". Outside researchers were not persuaded that one model will win. Pete Florence, cofounder of Generalist, said it is "unclear that any existing model would dominate robotics."

Reporting on the conference side is also disputed. The Decoder reports that OpenAI's booth was empty at the CoRL robotics conference in Austin despite being registered. A Bluesky post disputed that account.

## The Insider view

The clearest winners are Chen and Brockman, who now control research and commercial robotics between them, and any buyer who can access Astra through a general platform rather than a bespoke robot stack. The exposed party is any company selling specialised robotics models. On Fortune's figures, those models score 51.0% on the simulated benchmark, well below Astra's 74.8%.

The real-hardware numbers are the weak point. Seven successful tasks out of 100, at $36.40 per attempt, is far from a deployable product. Watch whether the benchmark is published, whether the 2027 humanoid demo happens on schedule, and whether the cost per attempt falls.

## Key takeaways

- OpenAI is dropping a standalone robotics bet in favour of general models.
- Research staff go to Mark Chen; hardware and applied robotics go to Greg Brockman.
- Reported cost of $36.40 per robot attempt sets the bar for rivals.

## Sources

- [OpenAI Disbands Standalone Robotics Team, Moves Researchers to Main Research and Applied Units](https://fortune.com/2026/11/13/openai-disbands-robotics-team-as-its-general-models-start-running-robots/) — Fortune, 2026-11-13
