Exclusive: OpenAI disbands its robotics team as its general models start running robots
Reported by Fortune, Breakingthenews.net, MLQ.ai and 2 more outlets5 sources ↓
OpenAI has disbanded its standalone robotics team and reassigned its researchers to other groups. The company told Fortune that tests of its general-purpose models on robotics tasks prompted the change.
Key points
- Robotics researchers moved to OpenAI’s main research team; applied robotics and hardware staff moved to its Applied organization
- OpenAI says general-purpose models, including GPT-6 Astra, can be applied directly to robotics
- Nvidia’s simulated benchmark recorded a 74.8% success rate for an Astra-powered agent, versus 51% for the best specialized models reported by Fortune
- OpenAI says it remains committed to robotics; a first public humanoid demonstration is targeted for 2027, Fortune reports
Sources
- FortuneRead ↗Exclusive: OpenAI disbands its robotics team as its general models start running robots
- Breakingthenews.netRead ↗OpenAI disbands standalone robotics team
- MLQ.aiRead ↗OpenAI disbands its standalone robotics team as its general models start running robots, Fortune reports
- The DecoderRead ↗OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessary
- Tech TimesRead ↗OpenAI Robotics Team Reorganized; GPT-6 Astra Tops Robot Models in Simulation, Takes 11-19 Seconds
Details
Fortune reports that OpenAI confirmed the reorganization after it had already booked a booth at the Conference on Robot Learning in Austin. The booth was empty when Fortune visited. OpenAI said robotics remains part of its long-term vision, and The Decoder described the move as folding researchers into other teams rather than abandoning robotics.
OpenAI told Fortune that its researchers initially expected to build on work such as its Sora video model, but later tests showed its general-purpose frontier models could be applied directly to robotics. In a simulated manipulation benchmark, an Astra-powered agent achieved a 74.8% success rate; Fortune compared that with 51% for the best specialized models. TechTimes reports that Nvidia measured Astra’s decision time at 11.50 to 18.56 seconds per call, depending on the setup, and cautions that the simulation did not establish real-world performance.
Written from coverage as of
Images
- Fortune
- The Decoder
- Tech Times
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh