🔍 Read the full analysis: Why OpenAI's Robotics Reorganization Has People Asking About Robot AI on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI has disbanded its stand-alone robotics team, moving researchers into its main research organization and applied robotics and hardware staff into its Applied organization. The company says general-purpose models can now be used directly for robot control, but published tests show a trade-off: general models performed strongly in simulation while a specialist model acted much faster.
OpenAI has broken up its stand-alone robotics team and moved its researchers into other parts of the company, a change it confirmed to Fortune. The company says its general-purpose models can now be applied directly to robot control, but available tests point to a practical trade-off: general models can perform well on simulated tasks, while specialist robot models respond much faster.
According to Fortune, OpenAI’s robotics researchers now report to Chief Research Officer Mark Chen, while its applied robotics and hardware teams have moved to the Applied organization, led by President Greg Brockman. The change was visible at the CoRL robotics conference in Austin: OpenAI’s booth remained on the floor plan, but was empty when Fortune visited. OpenAI confirmed the reorganization to the outlet; it had not published its own announcement as of November 14.
OpenAI’s spokesperson described the move as a sign of continued commitment, not a retreat from robotics. The company told Fortune that frontier models, including GPT-6 Astra, have shown they can be applied to robotics directly. Earlier in the year, researchers expected to build on video models such as Sora. The spokesperson said the newer results prompted the company to place robotics researchers alongside teams working on general models.
One cited example is a simulation study by NVIDIA researchers, published October 14 as a CoRL 2026 late-breaking result. On the RoboLab-120 benchmark of 120 simulated tabletop tasks, an agentic GPT-6 Astra setup achieved a reported 74.8% success rate, compared with 51.0% for the best trained robot policy listed in the study’s comparison. That result came with a substantial speed limitation: in the direct setup, each model call averaged 18.56 seconds, and the simulation paused while the model produced its response.
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 Trade-Off Between Skill and Speed
The reorganization puts a strategic question at the center of OpenAI’s robotics work: should a robot rely on a broad model that can interpret varied situations, or a specialist model built to produce actions quickly? Simulation performance alone does not settle that question. A system that succeeds at more tasks may still be unsuitable for time-sensitive physical work if it takes too long to decide what to do.
A different approach appeared two days before the CoRL report. Runway released Praxis-1, a 27-billion-parameter “world action model” with weights and code available under the Apache 2.0 license. Runway reports a 46.2% score on RoboLab-120, the highest among the world action models and open-source robot models listed there, and fifth overall. Its model takes about 0.55 to 0.6 seconds per call for 1.6 seconds of motion, according to the source material.
Runway also reported an internal comparison on one of its robots: GPT-6 Astra took 15.6 times longer than Praxis-1 to complete three tasks and cost 51 times more per completed task. Those figures come from Runway’s own test and should not be treated as an independent, broad comparison. Runway calls the approaches complementary, suggesting a possible division between general models that reason across tasks and specialist models that handle fast, repeated actions.
For readers, the distinction matters because robotics depends on more than whether a model can choose a plausible movement. Response time, reliability and operating cost affect whether a system can be used outside a paused simulation. The published results describe different setups and do not establish which approach will work best across real-world robots and tasks.
Products Worth Considering

Kindle Paperwhite 16GB (2024 Model) – Graphite

LG UltraGear 32GX870A-B 32-inch 4K UHD OLED Gaming Monitor

Bose QuietComfort Headphones – Wireless Bluetooth Over-Ear Noise Cancelling Headphones, Petal Pink Limited Edition
As an affiliate, we earn on qualifying purchases.
OpenAI’s Robotics Team, Rebuilt
This is the second time OpenAI has closed a stand-alone robotics team. The company wound down its first robotics effort in 2020. Cofounder Wojciech Zaremba later cited limited robot data and a more efficient path to artificial general intelligence through other methods. OpenAI rebuilt a robotics team in 2024 and partnered with humanoid maker Figure. Figure said in February 2025 that it would develop its own robot AI, ending the partnership.
OpenAI has said it is developing a humanoid robot of its own. CEO Sam Altman has targeted a first public demonstration in 2027. That goal is separate from the reported reorganization: moving teams within the company does not, by itself, confirm a change to the robot’s development schedule or design.
Fortune also described pressure on OpenAI to rein in costs and focus on core products. That is business background, not the explanation the company gave for the team changes. Its stated reason was that general-purpose frontier models had become applicable to robotics directly. OpenAI has not issued a public statement about the reorganization in its own news feed, according to the source material checked November 14.
“The two approaches are “complementary.””
— Runway
What the Tests Do Not Show
The available figures do not establish how well either approach performs across real-world robots, varied environments or sustained operation. NVIDIA’s reported Astra result comes from a simulation benchmark, and its setup paused the simulation while the model was thinking. The 74.8% result therefore should not be read as evidence that a physical robot can complete tasks at a comparable speed or reliability.
The comparisons also use different systems and conditions. Runway’s speed and cost figures come from its own test on one of its robots, while the benchmark scores come from RoboLab-120. The source material does not provide enough detail to treat those figures as a controlled, independent comparison across all models. It also does not specify the cost methodology behind the reported 51-fold difference.
OpenAI has not publicly detailed which robotics projects or staff changed, whether the company will continue building its own robot, or how it plans to combine general models with dedicated control systems. The 2027 demonstration target remains a stated goal; the reorganization does not confirm whether that timing has changed.
Evidence Needed Beyond Simulation
The next useful milestones will be independent tests on physical robots that report success, response time, reliability and cost under comparable conditions. Further results from the RoboLab-120 study and assessments of Runway’s Praxis-1 may help clarify how much of the simulation performance carries over to practical use.
OpenAI has not announced a next public update on the reorganization. Its stated longer-term marker is a public demonstration of its humanoid robot targeted for 2027. Until the company provides more detail or physical-system results emerge, the clearest confirmed picture is that OpenAI has moved its teams and is betting more directly on general models, while the speed case for specialist robot AI remains visible in the available comparisons.
Key Questions
What did OpenAI change in its robotics organization?
According to Fortune, 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. OpenAI confirmed the change to Fortune.
Why did OpenAI reorganize the teams?
OpenAI’s spokesperson told Fortune that general-purpose frontier models, including GPT-6 Astra, have shown they can be applied directly to robotics. The company described robotics as part of its long-term vision, not a discontinued effort.
Did GPT-6 Astra outperform specialist robot models?
In one NVIDIA simulation study, an agentic GPT-6 Astra setup achieved 74.8% success on RoboLab-120, above the 51.0% result reported for the best trained robot policy in the comparison. That setup averaged 18.56 seconds per call in the direct configuration, so the score does not establish faster real-world control.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter world action model released with weights and fine-tuning code under the Apache 2.0 license. Runway reports a 46.2% RoboLab-120 score and call times of about 0.55 to 0.6 seconds for 1.6 seconds of motion.
Has OpenAI canceled its humanoid robot plans?
The source material does not report a cancellation. OpenAI has said it is developing a humanoid robot, and Sam Altman has targeted a first public demonstration in 2027. Whether the reorganization affects that project or schedule remains unclear.
Source: ThorstenMeyerAI.com
