🔍 Read the full analysis: OpenAI's Robotics Team Split Raises A Big Question: Do Robots Need Their Own AI? on ThorstenMeyerAI.com
Get hardware and tech essentials delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI has disbanded its stand-alone robotics team and reassigned researchers across its research and Applied organizations. The company says general-purpose models can now be used directly for robot control, but available tests show a trade-off: general models perform well on simulated tasks, while specialist models act much faster.
OpenAI has broken up its stand-alone robotics team and moved its researchers into other parts of the company, a reorganization the company confirmed to Fortune. OpenAI says general-purpose models such as GPT-6 Astra can now be applied directly to robot control, even as published tests point to a practical trade-off: broad models can perform well in simulation, while specialist models respond faster.
According to Fortune’s November 13 report, OpenAI’s robotics researchers now sit within the main research organization under Chief Research Officer Mark Chen. Applied robotics and hardware teams have moved to the company’s Applied organization, which reports to President Greg Brockman. Fortune reported that it confirmed the changes with an OpenAI spokesperson; OpenAI had not published a separate announcement as of November 14.
The spokesperson described the reorganization as a sign of continued investment, not a retreat. Fortune quoted the company as saying robotics “remains a core part of OpenAI’s long-term vision” and that it remains committed to bringing general-purpose robots into the world. The company’s explanation is that frontier models, including GPT-6 Astra, have shown they can be used for robotics directly, changing the rationale for maintaining a separate research team.
A recent NVIDIA Robotics Lab simulation study offers one piece of evidence behind that approach. On RoboLab-120, a benchmark of 120 tabletop tasks for a simulated robot arm, an agentic GPT-6 Astra setup achieved a reported 74.8% success rate. The study said that result exceeded the best policy on the benchmark leaderboard at the time. In the direct setup, however, each model call took an average of 18.56 seconds, and the simulation paused while the model processed its response.
Runway has taken a different route. Two days before the CoRL conference visit described in the source report, it released Praxis-1, a 27-billion-parameter model for robot actions, with weights and code under the Apache 2.0 license. Runway reported a 46.2% score on RoboLab-120 and a typical call time of about 0.55 to 0.6 seconds for 1.6 seconds of motion. In Runway’s own test on one of its robots, the company said GPT-6 Astra took 15.6 times longer than Praxis-1 to complete three tasks and cost 51 times more per completed task. Those comparisons are Runway’s results, not an independent evaluation.
Robot control loop · explainer
Who decides the robot’s next move?
Every robot repeats one loop: cameras see, a model decides, the arm moves. The two bets in this story put a different kind of model into the decide step.
Option 1
A general model in the loop
Example: GPT-6 Astra in NVIDIA’s simulation study
18.56 s per model call on average in the direct set-up. Inference was 97.0% of the estimated episode time; the simulation paused while the model was thinking.
Option 2
A specialist robot model in the loop
Example: Runway’s Praxis-1, a world action model with open weights
About 0.55–0.6 s per call on one H100 server, and each call returns 1.6 s of motion. Needs fine-tuning data for each robot.
General model plans the next stepslow, outside the loop
Option 3 · illustrative
Both, in two layers
The idea Runway and the NVIDIA researchers both describe
Runway calls the two approaches “complementary”: an agent plans, a fast policy carries out each step. NVIDIA’s authors say the best way to combine them “remains an open question”.
Sources: NVIDIA Seattle Robotics Lab, “Astranomy” study of GPT-6 Astra, Oct 14, 2026 (simulation, direct set-up); Runway, Praxis-1 technical report, Nov 2026 (one H100 server). Loops and timelines are schematic; the two figures come from different set-ups, not from a head-to-head test.
Graphic: Powered by Thorsten Meyer AI
The Speed Gap in Robot Control
The reorganization puts a strategic question at the center of robotics development: can a general-purpose model handle physical tasks well enough that companies no longer need a separate model-building effort? OpenAI’s move suggests it sees value in connecting robotics more closely to its broader model research. It does not, on its own, establish that general models are ready to control robots reliably in everyday settings.
The reported results also show why the answer may not be a simple choice between general and specialist AI. The NVIDIA study measured task success in a simulation that stopped while the model generated an answer, while Runway’s reported timings point to the responsiveness needed for ongoing movement. Success rate, response time, cost and real-world reliability are different measures. A model that performs well on a simulated task may still be too slow or costly for a particular use.
For researchers and companies, this affects how robotics systems are built and evaluated. One possible arrangement is to use a broad model for interpreting instructions or planning, alongside a faster specialist model for immediate actions. Runway has described the approaches as complementary, but that is the company’s view; the cited results do not establish a standard design for future robots.
Products Worth Considering
LG UltraGear 24-Inch Full HD IPS Gaming Monitor, 120Hz (144Hz Overclocked), 1ms MBR, G-SYNC Compatible, FreeSync, HDR10
Trezor Safe 7 Crypto Hardware Wallet with Bluetooth and Color Touchscreen
Ledger Flex Crypto Wallet with 2.8-Inch E Ink Touchscreen
As an affiliate, we earn on qualifying purchases.
OpenAI’s Robotics Team, Rebuilt
This is the second time OpenAI has ended a stand-alone robotics effort. The company wound down an earlier team in 2020. It rebuilt a robotics team in 2024 and partnered with humanoid robot maker Figure; Figure said in February 2025 that it would develop its own robot AI, ending that partnership.
At CoRL, an academic robotics conference held in Austin during the week of the report, OpenAI’s booth remained listed on the floor plan but was empty when Fortune visited. The detail reflects how recently the team change had occurred; it is not itself an explanation for the decision. Fortune also reported that OpenAI is under pressure to control costs and focus on core products, but the company’s stated reason for the reorganization was the progress it sees in applying general-purpose models to robotics.
OpenAI has said it is developing a humanoid robot, and Fortune reported that CEO Sam Altman has targeted a public demonstration in 2027. The company has not released a public statement detailing the team changes or how the new structure connects to that project.
“The two approaches are “complementary.””
— Runway
What the Benchmarks Cannot Show
The published figures do not establish how either approach performs across a range of physical robots, environments and longer tasks. The NVIDIA result comes from a simulation benchmark, while Runway’s speed and cost comparison comes from the company’s own test on one of its robots. The available information does not provide an independent, directly comparable evaluation of Praxis-1 and GPT-6 Astra under the same conditions.
It is also unclear exactly how OpenAI’s redistributed teams will work together, whether the company will publish more detail about its robotics plans, or how it will address the latency and cost shown in the cited tests. OpenAI had not posted its own announcement about the change as of November 14, according to the source report.
The 2027 Robot Demonstration
The next public milestone identified in the reporting is OpenAI’s previously stated target of a humanoid robot demonstration in 2027. The company has not provided further details in the material available about the robot, its capabilities or whether it will use GPT-6 Astra directly.
Further evidence will depend on published evaluations and demonstrations that clarify how models perform outside simulation, including their speed, cost and reliability during physical tasks. For now, OpenAI’s reorganization confirms a change in how it organizes robotics work, not that general-purpose models have replaced specialist robot AI.
Key Questions
What changed in OpenAI’s robotics organization?
Fortune reported that OpenAI moved its robotics researchers into the main research organization under Mark Chen and its applied robotics and hardware teams into the Applied organization under Greg Brockman. The company confirmed the change to Fortune.
Why did OpenAI reorganize its robotics team?
According to an OpenAI spokesperson quoted by Fortune, the company’s general-purpose frontier models, including GPT-6 Astra, have shown they can be applied directly to robotics. OpenAI framed the move as a change in approach, not a retreat from robotics.
Does GPT-6 Astra outperform specialist robot models?
In an NVIDIA simulation study, an agentic GPT-6 Astra setup scored 74.8% on RoboLab-120. But the direct setup averaged 18.56 seconds per model call. Runway reported faster calls for its specialist Praxis-1 model. The results use different setups and do not establish a universal winner.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter model for robot actions. Runway released its weights and code under the Apache 2.0 license and reported a 46.2% RoboLab-120 score.
What remains unknown about OpenAI’s plans?
OpenAI has not publicly detailed how the reorganized teams will develop robotics systems or how its general models will be used in physical robots. A public humanoid demonstration has been reported as a 2027 target, but further project details were not provided.
Source: ThorstenMeyerAI.com
