Picture a robotics conference where OpenAI has registered a booth, the floor plan lists it, and the booth is empty. According to Fortune, that is what its reporter found at booth #208 at CoRL, a top academic robotics conference in Austin, this week, with no OpenAI logo in sight. The reason, Fortune says, is a reorganization that happened recently enough that the booking was already made.
On 13 November 2026, Fortune reported that OpenAI has disbanded its stand-alone robotics team and reassigned the researchers. The company’s pitch is that this is about speeding robotics up, because it now thinks one general-purpose model can do much of the job a dedicated robot brain used to do. Here is what is known, what the evidence is, and where the doubters push back.
What happened
Fortune’s exclusive, by Wen Shao, says OpenAI confirmed the reorganization when asked. Per that report, the robotics research team now sits in the main research group under Chief Research Officer Mark Chen. The applied robotics and hardware teams moved into the larger Applied organization, which reports to President Greg Brockman.
An OpenAI spokesperson told Fortune that robotics remains a core part of the company’s long-term vision and that it is still committed to bringing general-purpose robots into the world. Fortune also reports that Sam Altman is targeting a first public demonstration of OpenAI’s own humanoid robot in 2027.
This is not the first reshuffle. Fortune notes that OpenAI wound down its original robotics team in 2020, with cofounder Wojciech Zaremba later citing limited robot data. It rebuilt a stand-alone team in 2024 alongside a partnership with the startup Figure, which ended in February 2025 when Figure said it would build its own robot AI.
The argument: robots as one more thing a smart model does
OpenAI’s explanation, as relayed by Fortune, is that at the start of the year its researchers assumed the best route to robotics was building on work such as Sora, its video model, which has to learn something about how the physical world behaves. Since then, the spokesperson said, its general-purpose frontier models, including GPT-6 Astra, have shown they can be applied directly to robotics.
OpenAI launched Astra on 3 September 2026, calling it “the world’s most intelligent and aligned model.” Guanya Shi, an assistant professor at Carnegie Mellon University’s Robotics Institute, told Fortune the move signals OpenAI sees robotics as “no longer a special problem” and instead as a downstream use of general intelligence, much like coding or legal tools built on the same underlying models.
It is a striking turn for a company that, in 2019, announced a robot hand solving a Rubik’s Cube using neural networks trained entirely in simulation. OpenAI highlighted that it coped with surprises such as being prodded by a stuffed giraffe. The new bet is that a purpose-built system may not be needed at all.
What the numbers say (and don’t)
Fortune cites a study by Nvidia in which an Astra-powered agent was given camera images, robot information and a control interface. It reached a 74.8% success rate on a simulated manipulation benchmark, compared with 51.0% for the best specialised robot models on that leaderboard. Fortune stresses that this was a digital simulation, not a physical robot. We could not find the Nvidia study itself, so that figure rests on Fortune’s reporting.
A separate study gives a more mixed picture. Neurohive’s write-up, dated 19 October 2026, of work by the SenseNova team and university partners says Astra was tested on 12 simulation benchmarks and 8 real robots without task-specific fine-tuning. It reportedly beat the best published results on 9 of the 12 benchmarks.
The same write-up lists the caveats:
- Some results depended on a larger step budget, up to 10 times what the benchmark rules allow.
- Even a nearly solved task took about an hour per episode.
- On real robots, the model struggled with precise contact and reportedly never managed to insert a plug into a socket.
Because these studies use different setups, their numbers should not be compared with each other.
Not everyone is convinced
Shi, quoted by Fortune, still sees a need for smaller specialised models that run on a robot’s own computer. Large language models are powerful but slow and need to be hosted in the cloud, so a robot would need a constant network connection. If you have ever run a chatbot on your own laptop, our guide to running an AI model with Ollama shows the trade-off in miniature.
Russ Tedrake, an MIT professor and cofounder and CEO of Walden Robotics, told Fortune that robotics companies keep an edge the AI labs lack. Their robots already work in real places such as factory floors, and the data from that “deployment” helps robots adapt to specific workplaces. Better base models, he said, make that data more valuable.
Pete Florence, cofounder of Generalist, said it is unclear that any existing model will dominate robotics in the long run. Fortune adds that the one-brain idea would seem to challenge the premise behind many robotics startups that have raised billions on the belief that robots need brains of their own.
The money angle
Fortune also puts the move in a business context. It says OpenAI is under pressure to rein in costs and focus on a few core products. Last year, it says, Altman told the company to abandon “side quests”, and the company has dropped some efforts, including Sora, to free up computing power. Fortune adds that OpenAI’s operating losses reportedly totalled $10.9 billion in the most recent quarter as it seeks a new funding round and prepares for a possible IPO in early 2027.
OpenAI’s own explanation is about acceleration, and the cost angle is Fortune’s framing. Both can be true at once.
What it means for you
Fortune’s report does not describe any change to ChatGPT itself. The signal is about direction: the same family of models you use for writing and coding is being tested as a robot controller. The same theme runs through products like OpenAI’s agents that can now pay for you, which we covered in our ChatGPT Wallet explainer.
As of 15 November 2026, the real test is still ahead. That means robots working outside simulations, at a speed and cost that make sense, and the humanoid demo Fortune says Altman is aiming for in 2027.
Sources
Links last checked on 15 November 2026.
- Exclusive: OpenAI disbands stand-alone robotics team as general AI models start running robots, Fortune (13 November 2026)
- GPT-6 Astra: A new generation of intelligence, OpenAI (3 September 2026)
- Solving Rubik’s Cube with a robot hand, OpenAI (15 October 2019)
- GPT-6-Astra Has Crossed the Threshold of Embodied Competence, but Not Embodied Intelligence: A SenseNova Study, Neurohive (19 October 2026)





Write a response