Key Insights
- A paradigm shift: OpenAI’s robotics focus is shifting more and more toward using general‑purpose AI to build robots than focusing on robotics as a separate field.
- A challenge: The simulation report shows a score of 74.8 percent. This supports the argument for using foundation models. Yet this score does not give an advantage, for real-life deployments.
- One part of the budget problem: OpenAI is restructuring in the wake of its proposed 2027 filing to go public and cut spending.
OpenAI has dissolved its standalone robotics team just five months after rebuilding the unit. The company, which has recently seen GPT-6 Astra show its robot-control prowess, is bringing in researchers to its AI core businesses.
🚨BREAKING: OpenAI has disbanded its robotics team, just 5 months after relaunching it.
The company has moved its robotics researchers into its main AI lab, telling Fortune that GPT-6 Astra can already control robots across certain tasks.
The move comes after robotics chief… pic.twitter.com/YaDGj4Aj39
— Coin Bureau (@coinbureau) November 14, 2026
The move comes after Caitlin Kalinowski, OpenAI’s robotics chief, stepped down due to the company’s deal with the Pentagon. She spoke of concerns with domestic surveillance and lethal decisions made without human authorization, and that these issues needed to be looked at in more detail. Her leaving was a matter of conscience, she explained.
OpenAI Restructures Its Robotics Division
OpenAI confirmed the reorganization following a report by Fortune. Robotics researchers will now report to Chief Research Officer Mark Chen.
Meanwhile, applied robotics and hardware specialists will join the Applied organization under President Greg Brockman. OpenAI maintains that robotics remains central to its long-term strategy.
The changes came shortly before the Conference on Robot Learning in Austin. OpenAI had registered booth 208, but visitors found the space empty without company branding or demonstrations.
GPT-6 Astra Drives a New Research Direction
OpenAI previously ran a robotics team. It shut that team down in 2020. The cofounder Wojciech Zaremba said that the amount of robot data was limited and that stronger opportunities lay in language models.
The firm then shifted its focus to big AI models, and it finally released Chat GPT back in 2022. In 2024, OpenAI, however, reconstructed its robotics teams and formed a collaboration with Figure to create a humanoid robot technology.
In February 2025, the partnership was ended, with Figure announcing it would construct its own AI models in its stead. OpenAI continued developing a humanoid robot, with Sam Altman targeting a public demonstration in 2027.
Recently scientists looked at models like Sora to learn how things move. However OpenAI said that GPT‑6 Astra has potential, for world robot tasks. An Nvidia simulation study found a 74.8% success rate, for an agent using Astra. In comparison specialized robotics models got 51% on the test.
The test gave camera images, robot data and a control panel. It also did not need any training, on robotic data sets.
General AI Challenges Specialized Robotics Models
The restructuring shows that people are becoming more interested, in AI. This approach uses general-purpose models to control robots, autonomous vehicles and drones.
Guanya Shi of Carnegie Mellon’s Robotics Institute said the move suggests OpenAI sees robotics as an application of general intelligence.
He noted that improved spatial reasoning helps models handle changing environments. Improved spatial reasoning allows models to recognize objects adapt to lighting changes and identify mistakes during tasks.
However, specialized robotics companies retain advantages. MIT professor Russ Tedrake said that robotics companies’ machines already collect valuable data from factories and other workplaces.
That experience helps robots adapt to environments. Stronger foundation models could also make deployment data more useful.
Generalist cofounder Pete Florence questioned whether existing model would dominate robotics. Pete proposed options that may lead to progress in the future.
The limitations of cloud-based AI were also mentioned by Shi. The larger the model, the slower it runs and the more it demands of a connection to the internet. Instead, smaller specialized models can be implemented directly on a computer onboard.
The priorities of OpenAI are influenced by financial pressures.
OpenAI is also under pressure to cut costs and focus on the more profitable products. Sam Altman has been urging workers to shun side projects and focus on the main line of business.
The company has cut down on other programs, such as Sora, to free up computer resources. It is also exploring funding and is currently considering an IPO in early 2027.
In its most recent quarter, it reported operating losses of some $10.9 billion. With this financial constraint, better use of resources becomes a more pressing concern.
OpenAI’s decision is a concern for robotics startups going forward. Billions of dollars have been raised by various companies to develop AI systems for industrial and humanoid robots.
But with more sophisticated general models growing in their capabilities, that business model might be endangered.
Conclusion
OpenAI is redirecting its robotics research into its wider AI endeavors, don’t worry, it isn’t stopping the work at the robotics end. While this approach is backed by the reported performance of GPT-6 Astra, it is still a challenge to put it into practice.
However, there are benefits to specialized models in terms of speed, connectivity and real-world experience. The next big challenge in the industry will be seeing if general AI can do those things.





