
Key points
- OpenAI has dissolved its standalone robotics team, moving researchers to main research and applied organisations.
- An Nvidia study reported GPT-6 Astra achieved a 74.8 percent success rate on simulated tasks, outperforming specialized models.
- The company frames the restructuring as a move to accelerate robotics efforts using general-purpose model capabilities.
OpenAI is restructuring its robotics operations by dissolving its standalone team and redistributing researchers across other departments, according to a report by The Decoder.reported The company states this move aims to accelerate its robotics initiatives rather than abandon the field.The company states this move aims to accelerate its robotics initiatives rather than abandon the field.The company states this move aims to accelerate its robotics initiatives rather than abandon the field.
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
The evidence
General-purpose models like GPT-6 Astra appear capable of controlling robots directly, reducing the perceived need for specialized robotics models.reported An Nvidia study reported that an Astra-powered agent achieved a 74.8 percent success rate on simulated robotics tasks involving a DROID robot.reported This figure compares to a 51 percent success rate for the best specialized models in the same tests.reported
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
A separate benchmark called StationeryBench compared GPT-6 Astra with Ai2's MolmoAct2 on five tasks using dual-arm YAM robotic arms.reported Across 200 trials, Astra fully completed seven out of 100 tasks, while MolmoAct2 completed none.reported Astra also recorded an average progress score of 46, compared to 12 for MolmoAct2, and demonstrated navigation and cleaning abilities in a humanoid robot kitchen test.reported
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
The decision
Robotics researchers are moving to the main research team under Chief Research Officer Mark Chen, while applied robotics and hardware teams join Greg Brockman's Applied organization. The restructuring occurred with such short notice that OpenAI had already booked a booth at the Conference on Robot Learning in Austin, which remained empty, Fortune reports.reported
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
The company rebuilt its robotics team in 2024 after shutting it down in 2020 due to a lack of training data.reported CEO Sam Altman is reportedly targeting a first public robot demo for 2027, with long-term goals to build everyday robots.reported
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
The takeaway
It remains unclear how much robotics data OpenAI and others incorporated during training for these models. Other large general-purpose models from Anthropic and Google have also made recent progress on robotics tasks. Specialized robotics models may still hold advantages in cost, reliability, speed and energy consumption, even if they lack the generalization capabilities of newer models.
Building on this
Developers can leverage GPT-6 Astra’s documented spatial understanding and task completion rates for simulated robotics applications.reported The model’s ability to control simulated DROID robots and navigate humanoid platforms suggests potential for integration into testing environments. However, the lack of clarity on training data composition means developers must verify performance in physical deployments. The superior progress scores in StationeryBench indicate Astra may serve as a viable control layer for complex object manipulation tasks, provided latency and energy constraints are managed.
Evidence: reported
OpenAI reassigns robotics researchers to other teams as general-purpose models might make specialized robotics AI unnecessaryaifleet-decoder
Why it matters
- Enterprise IT buyers should monitor whether general-purpose models reduce the need for specialized robotics software stacks, potentially simplifying procurement.
- The shift in OpenAI’s internal structure signals that industry leaders view direct model control as a viable path, influencing future vendor strategies.
- Developers must weigh the higher success rates of general models against potential costs and energy consumption compared to specialized alternatives.
- Uncertainty around training data composition requires caution when deploying these models in safety-critical or high-reliability physical environments.
Topics: OpenAI, robotics, GPT-6 Astra, AI models, benchmarks
Updated
For readers
Developers
- Consider using GPT-6 Astra for simulated robotics tasks where high progress scores are needed.
- Verify performance in physical deployments due to uncertainty around training data composition.
Business
- Monitor shifts in vendor strategies as general models reduce the need for specialized robotics units.
- Evaluate cost and energy tradeoffs between general and specialized models for enterprise applications.
Researchers
- Investigate the impact of 3D training data on spatial understanding in general models.
- Compare general model performance against specialized models in reliability and speed metrics.
Ask the article
Local keyword search of this article; no generated answers.Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Terms in this story
GPT-6 Astra
A general-purpose model from OpenAI capable of controlling robots directly.
StationeryBench
A robotics benchmark comparing model performance on desk object tasks.
DROID robot
A simulated robot used in Nvidia study tests for robotics tasks.
MolmoAct2
A specialized robotics model from Ai2 used for comparison in benchmarks.
Questions readers ask
Why is OpenAI dissolving its robotics team again?
The company believes general-purpose models like GPT-6 Astra can control robots directly, making a dedicated specialized unit unnecessary.
How did GPT-6 Astra perform in robotics benchmarks?
It achieved a 74.8 percent success rate in an Nvidia study and outperformed MolmoAct2 in StationeryBench tasks.
Are there any downsides to using general models for robotics?
Specialized models may still offer advantages in cost, reliability, speed and energy consumption.
Connected stories
Reading receipt
- OpenAI has dissolved its standalone robotics team, moving researchers to main research and applied organisations.
- An Nvidia study reported GPT-6 Astra achieved a 74.8 percent success rate on simulated tasks, outperforming specialized models.
- The company frames the restructuring as a move to accelerate robotics efforts using general-purpose model capabilities.
Still open: How much robotics data was incorporated into the training of GPT-6 Astra?
Still open: Will specialized robotics models remain competitive in cost and reliability against general models?