OpenAI disbanded its independent robotics team and shifted to using general-purpose models to control robots.
BIBIBI
AT A GLANCE
According to the report, OpenAI has merged robotics research into other teams and shifted to using general AI models to control robots.
Article
OpenAI shifts to using general AI models to directly control robots, no longer developing separate specialized models
BlockBeats reports that on November 14, according to Fortune magazine, OpenAI has disbanded its independent robotics research team and reassigned the relevant researchers to other departments within the company. OpenAI stated that this move does not mean abandoning the robotics business, but rather aims to accelerate the development of general-purpose robots. Following the adjustment, robotics researchers will be integrated into the core research team led by Chief Research Officer Mark Chen, while the application robotics and hardware teams will be transferred to the applications division overseen by President Greg Brockman.
This restructuring reflects a shift in OpenAI's strategic direction: the company believes that general AI models can now be directly applied to robot control, eliminating the need to develop dedicated standalone robot AI models. OpenAI claims that cutting-edge general models, including GPT-6Astra, have already demonstrated capabilities for robotic applications. A study by NVIDIA shows that agents powered by Astra achieved a success rate of 74.8% in simulated robotic manipulation tests, surpassing the previously reported 51% for dedicated robot models on that leaderboard.
OpenAI still plans to develop humanoid robots, with CEO Altman previously aiming to conduct the first public demonstration in 2027. However, industry insiders point out that general models still face limitations such as slower operating speeds and reliance on cloud networks, meaning robotics companies retain advantages in real-world deployment and data accumulation.
BlockBeats, citing Fortune, reports that OpenAI has disbanded its independent robotics research team and reassigned the relevant researchers to other departments within the company.
02
The report says that robotics researchers will be merged into the core research team led by Marc Chen; the applied robotics and hardware teams will be transferred to the applications division headed by Greg Brockman.
03
The report notes that OpenAI believes general AI models can be used directly for robot control, eliminating the need to develop separate robot AI models; OpenAI states that cutting-edge general models, including GPT-6Astra, have already demonstrated capabilities for robotic applications.
04
The report cites an NVIDIA study stating that agents equipped with Astra achieved a 74.8% success rate in simulated robotic manipulation tests, surpassing the 51% reported previously for dedicated robotic models on this leaderboard.
05
The report says that OpenAI still plans to develop humanoid robots; Altman previously stated he aims to conduct the first public demonstration in 2027.
06
The report noted that industry insiders pointed out that general-purpose models have limitations such as slow operating speeds and reliance on cloud networks, while robotics companies still hold advantages in real-world deployment and data accumulation.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
This report describes organizational and technical roadmap adjustments in OpenAI's robotics research: relevant personnel have been integrated into other teams within the company, and efforts are underway to have general-purpose models handle robotic control tasks. The success rates cited in the report come from simulation tests and do not mean that robots have achieved the same performance in real-world scenarios.
Why it matters to readers
If general-purpose models can be used for robot control, it could affect how robotics research is organized and its technological roadmap; however, the material does not provide data on actual deployment performance or direct financial and market response.
This can be understood as a shift from training models specifically for robots to attempting to have general AI models directly control robots. Performance in simulated environments cannot alone prove that real robots can also operate stably.
Risks and unknowns
The material is a report cited by BlockBeats from Fortune, without providing Fortune’s original report or OpenAI’s announcement; the claims regarding team adjustments and corporate strategy cannot be independently verified here.
The material does not provide the original NVIDIA research paper, testing methods, or sample details; therefore, the comparison between 74.8% and 51% cannot be independently verified, nor can it be used to assess real-world performance.
The speed, reliability, network dependency, and deployment cost of the general model in real robot scenarios are not specified.
Related Developments
Loading event timeline…
The first public demonstration in 2027 was a planned statement reported by the media, and does not indicate that the demonstration has been confirmed or completed.
Related concepts
General AI models: The report refers to models that can be directly used for robot control, rather than robot-specific models developed separately.
This term is not in the glossary yet. Browse related concepts in the glossary.
View glossary →Robotic operation simulation test: testing the robot's operational capabilities in a simulated environment; the material does not specify the exact testing method.
This term is not in the glossary yet. Browse related concepts in the glossary.