Published on

OpenAI Folds Its Robotics Team Into Research as GPT-6 Astra Takes On Robot Control

Get your news fromHumanoids Daily

One tap, and our reporting appears more often in your Google Top Stories. It makes a real difference to a small newsroom.

Pontus Armini
Written byPontus Armini

Supported by RoboStrategy

Sponsored

Investing involves risk, including possible loss of principal. Read the prospectus before investing.

OpenAI no longer has a stand-alone robotics team. Last week the company confirmed to Fortune that it had split the group up: its robotics researchers now sit in the main research organization under Chief Research Officer Mark Chen, and the applied robotics and hardware teams have moved into the Applied organization run by President Greg Brockman. OpenAI's stated reason is that its general-purpose models, GPT-6 Astra among them, have proved they can be applied to robots directly.

The change undoes a structure that was only six months old. On 31 May, CEO Sam Altman announced that the company's world simulation research program had evolved into OpenAI Robotics, with job listings for actuator designers, simulation engineers and a data-collection workforce.

  • OpenAI's robotics researchers now report to Chief Research Officer Mark Chen. The applied robotics and hardware teams moved to the Applied organization under President Greg Brockman.
  • OpenAI says its general models, including GPT-6 Astra, can be applied to robots directly. Early in 2026 its researchers expected to build on video models such as Sora.
  • OpenAI staff disputed reports that the team had been shut down. One pointed to 15 open robotics roles, and the company says robotics remains part of its long-term plan, including a humanoid.
  • Google made a similar move a week later, when its general Gemini 4 Hydrogen model took over from the dedicated Gemini Robotics ER 2.
A spiral galaxy of glowing blue, white and orange points around a bright core against a black sky.
OpenAI's promotional artwork for GPT-6 Astra. Credit: OpenAI.

What changed, and what did not

Fortune's report said the team had been disbanded and its researchers reassigned. OpenAI staff objected to that word within hours. "robotics at openai is alive and well," OpenAI engineer David Schnurr wrote on X on the evening of 13 November, linking to 15 open robotics roles on the company's careers page. Welton Wang, another OpenAI employee, replied to a post summarizing the story as a shutdown that it was "completely wrong," adding that the company is investing more than ever in robotics.

Fortune's report and the staff posts are not far apart on the facts. There is no longer a robotics group that sits apart from the rest of the company, while the people, the hardware work and the hiring continue inside OpenAI's two largest organizations. An OpenAI spokesperson told Fortune that robotics "remains a core part of OpenAI's long-term vision." OpenAI has not said how many people moved.

Follow humanoid robotics without following every announcement.

Get the week’s most important launches, funding and research—and why they matter—in one email from Humanoids Daily.

Once a week. Unsubscribe anytime.

Read recent issues

The hardware effort stays in place. Altman said in September that OpenAI "will definitely do a humanoid", alongside other robots for data centers, and Fortune reports he is targeting a first public demonstration in 2027. Moving the hardware team under Brockman puts it in the same organization as OpenAI's products.

The reorganization was visible at CoRL in Austin the same week. OpenAI had registered a booth at the conference. When Fortune visited, it had no logo, display or demo.

Why OpenAI changed course

When Altman announced OpenAI Robotics in May, he described it as an outgrowth of world simulation research led by Aditya Ramesh, who previously led the DALL-E models. According to OpenAI, its researchers assumed at the start of the year that the best route to robot intelligence ran through video models such as Sora. OpenAI has since dropped Sora to free up compute, one of the cuts that followed Altman's instruction to drop "side quests."

What changed the plan was GPT-6 Astra. Since its release on 3 September, researchers inside and outside OpenAI have connected the model to robots and given it camera images, robot state and a control interface. An OpenAI roboticist had it paint the Golden Gate Bridge with a robot arm. Third-party tests by RoboCurve put it at 95% on gross pick-and-place, though only two of 20 precision insertions succeeded. On a 200-episode MolmoSpaces subset it scored 70.5%, ahead of every open-source robot policy in the comparison, and it led SuperCLUE's September robot-brain evaluation with a wide margin in planning. Fortune also cites an NVIDIA study in which an Astra-based agent reached 74.8% on a simulated manipulation benchmark, against 51.0% for the best specialized robot models.

Unitree founder Wang Xingxing said in October that his own GPT-6 tests had made him more optimistic about embodied AI.

The other frontier labs

OpenAI is not alone in folding robotics into its general models. On 17 November Google said its cheaper Gemini 4 Hydrogen model beats Gemini Robotics ER 2, the embodied-reasoning model Google DeepMind built specifically for robots. A week earlier, Anthropic reported that Claude Fable 5.2 scored 64.9% on the RoboDojo two-arm simulation benchmark, against 29.8% for Astra in Anthropic's own runs.

Researchers have also been using these models to build the tools of robot learning: writing simulation environments, reviewing failed episodes and steering dedicated skill models such as NVIDIA's GR00T 2.

What specialized models still offer

Guanya Shi, an assistant professor at Carnegie Mellon's Robotics Institute, told Fortune that he reads the move as a sign OpenAI sees robotics as "no longer a special problem" but as an application of general intelligence. He added that smaller models running on the robot itself still matter, because frontier models are slow and run in the cloud.

Pete Florence, co-founder of Generalist, which trains general-purpose models for robots, said it is unclear whether any existing model will dominate robotics in the long run. Russ Tedrake, an MIT professor and co-founder and CEO of Walden Robotics, argued that robotics companies keep an advantage in data from robots deployed on factory floors.

Most of Astra's strongest numbers come from simulation, and on harder two-arm tasks, pairing it with a learned robot policy raised success from 26% to 48%. NVIDIA's demos at CoRL used Astra as a planner on top of GR00T 2 rather than as the controller.

How OpenAI got here

This is the third structure OpenAI has tried for robotics. It closed its original robotics team in 2020, after years of work that included a robot hand solving a Rubik's Cube. Co-founder Wojciech Zaremba later pointed to the shortage of robot data as one reason. In 2024 it rebuilt a team and partnered with Figure on models for Figure's humanoids. Figure ended that partnership in February 2025 to build its own AI, and CEO Brett Adcock later said Figure had "fired" OpenAI. In March 2026, OpenAI's hardware lead Caitlin Kalinowski resigned over a Pentagon contract, the same week former K-Scale Labs CEO Benjamin Bolte joined the company. Two months later came OpenAI Robotics.

The new arrangement comes as OpenAI tries to contain costs. Fortune reports operating losses of $10.9 billion in the most recent quarter, a search for new funding and a possible IPO in early 2027.

Share this article

Follow humanoid robotics without following every announcement.

Get the week’s most important launches, funding and research—and why they matter—in one email from Humanoids Daily.

Once a week. Unsubscribe anytime.

Read recent issues