OpenAI is reorganizing its robotics work, moving researchers into other parts of the company as general-purpose AI models show progress on robot tasks. The change does not mean the company is abandoning robotics: it says the move is intended to accelerate the work.
A team reshaped across the company
Robotics researchers are moving to the main research team under Chief Research Officer Mark Chen. The applied robotics and hardware teams are joining Greg Brockman's Applied organization.
The decision came after OpenAI had already booked a booth at the Conference on Robot Learning (CoRL) in Austin. The booth sat empty, Fortune reports. The episode suggests the reorganization happened on short notice, but the source does not explain the timing or the plans for the booth.
OpenAI apparently sees general-purpose models such as GPT-6 Astra as potential robot controllers. If one model can handle a range of physical tasks, the company may need fewer models built for individual robotics applications. That possibility is the rationale behind the reported shift, not proof that specialized systems are no longer useful.
Early results show promise, with limits
An Nvidia study reported a 74.8 percent success rate for an Astra-powered agent on simulated robotics tasks. The best specialized models scored 51 percent in the same comparison. The test used a simulated DROID robot, not a physical one, so the result does not establish how the model performs in a real environment.
A separate robotics benchmark, StationeryBench, compared GPT-6 Astra with Ai2's MolmoAct2 on five tasks involving desk objects and dual-arm YAM robotic arms. Across 200 trials, Astra fully completed 7 out of 100 tasks, while MolmoAct2 completed none. Astra's average progress score was 46, compared with 12 for MolmoAct2.
Those figures tell different parts of the story: Astra completed some tasks, but the completion count remained limited. The benchmark also measured progress, which gives a broader view of performance than completion alone. In another test, Astra running on a humanoid robot navigated and cleaned up a kitchen.
What the results do not settle
It remains unclear how much robotics data OpenAI and other companies used to train their models. That uncertainty makes it difficult to tell from the reported results how much of Astra's performance comes from general-purpose capabilities and how much may depend on training data related to robotics.
Other large general-purpose models from Anthropic and Google have also made recent progress on robotics tasks. Yet specialized robotics models may retain advantages in cost and reliability, as well as speed and energy consumption, even if they generalize less well. A broad model's ability to handle varied tasks and a specialized model's efficiency are different strengths; the source does not show that one has made the other obsolete.
Yoav Artzi, an AI researcher at Cornell University and Google DeepMind, sees Astra as a major leap in spatial understanding. According to Artzi, GPT-6 Astra reaches near-human accuracy on the still-unpublished REMAP benchmark. He suspects OpenAI may have trained the model on large amounts of 3D training data such as Blender scenes. That is Artzi's suspicion, while the amount and nature of the training data remain unclear.
A new chapter in OpenAI's robotics plans
OpenAI shut down its robotics division in 2020. Co-founder Wojciech Zaremba said the company lacked the training data it needed and that large language models were a faster path to AGI. The company rebuilt the team in 2024, making the latest reorganization another change in how it organizes its robotics effort.
CEO Sam Altman is reportedly targeting a first public robot demo for 2027, and OpenAI wants to build its own everyday robots in the long term. At the same time, the company faces heavy financial pressure to focus on its core AI business, where it has lost significant ground to Anthropic.
The reorganization therefore sits alongside both technical and business questions. General-purpose models may make it easier to bring capabilities across tasks, but simulated results and early benchmarks do not settle their reliability, cost, speed, or energy use in practical robotics. OpenAI's plans point to continued interest in robots, even as the teams responsible for that work are being folded into larger organizations.