🔍 Read the full analysis: OpenAI Splits Up Its Robotics Team: Is Robot-Specific AI The Next Step? 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 broken up its stand-alone robotics team, moving researchers into its main research group 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 available tests show a trade-off: stronger simulation performance for a general model and much faster operation for a specialist model.
OpenAI has disbanded its stand-alone robotics team and moved its researchers and applied robotics staff into other company divisions, Fortune reported November 13. The company told Fortune the change reflects progress in general-purpose models such as GPT-6 Astra, which it says can be applied to robot control, rather than a retreat from robotics.
According to Fortune, OpenAI moved its robotics researchers into the main research organization under Chief Research Officer Mark Chen. Its applied robotics and hardware teams moved to the Applied organization, led by President Greg Brockman. OpenAI confirmed the reorganization when Fortune asked, but had not published its own announcement as of November 14. The company’s booth at the CoRL robotics conference in Austin remained listed on the floor plan, but was empty when Fortune visited.
An OpenAI spokesperson told Fortune that general-purpose frontier models, including GPT-6 Astra, have shown they can be applied directly to robotics. The spokesperson described robotics as part of the company’s long-term vision and said OpenAI remains committed to bringing general-purpose robots into the world. The company has also said it is developing a humanoid robot; CEO Sam Altman has targeted a first public demonstration in 2027.
Research results cited in coverage point to both the potential and the limits of the approach. In a simulation study, NVIDIA researchers tested GPT-6 Astra on RoboLab-120, a benchmark of 120 tabletop tasks for a robot arm. An agentic setup achieved a 74.8% success rate, above the 51.0% score of the best trained robot policy listed on the benchmark. But in the direct setup, a model call took an average of 18.56 seconds, and the simulation paused while it processed the request. These are simulation results, not evidence of equivalent performance on a physical robot.
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 signals a change in how OpenAI is organizing its robotics work: instead of keeping a separate team, it is placing research and applied hardware work alongside groups building its broader models and products. That structure is consistent with the company’s stated view that general-purpose AI may be useful for controlling robots without requiring a separate robot-specific model for every task.
But task success is only one measure of a robot-control system. Robots operating in the physical world need timely, repeated decisions; a model that performs well in a paused simulation may not be suitable for responsive movement if each call takes many seconds. A separate specialist approach may offer a speed advantage even if it scores lower on the cited benchmark. The tests therefore do not settle whether one kind of model can replace the other.
The difference also affects cost and access. Runway released Praxis-1, a 27-billion-parameter robot-focused model, with weights and code under the Apache 2.0 license. In Runway’s own test on one of its robots, it 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 figures are from Runway’s test and should not be treated as a general comparison across robots or tasks.
Products Worth Considering
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
LG UltraGear 32GN600-B 32-Inch QHD Gaming Monitor
Logitech MX Master 3S Bluetooth Wireless Mouse, Graphite
As an affiliate, we earn on qualifying purchases.
OpenAI’s Robotics Path So Far
This is the second time OpenAI has closed a stand-alone robotics team. The company wound down its first team 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-robot maker Figure; Figure said in February 2025 that it would develop its own robot AI, ending the partnership.
The latest change comes as OpenAI shifts its stated technical rationale. Earlier this year, researchers expected to build on video models such as Sora, according to the spokesperson’s account to Fortune. The company now says general-purpose models can be applied to robotics directly. The move also arrives amid pressure on OpenAI to manage costs and focus on core products, a business backdrop Fortune noted. OpenAI’s stated reason for the reorganization, however, is the reported progress of its general models.
The debate is not limited to OpenAI. Runway released Praxis-1 two days before Fortune’s report, presenting a dedicated “world action model” with open weights and commercial-use licensing. Its reported 46.2% RoboLab-120 score placed it fifth overall and first among the world action models on that leaderboard. In the cited comparison, Praxis-1 took about 0.55 to 0.6 seconds per call to produce 1.6 seconds of motion, illustrating the different emphasis on responsiveness.
“Complementary.”
— Runway, describing Praxis-1
What the Tests Cannot Show
The cited GPT-6 Astra results come from a simulation benchmark. They do not establish how the system performs on physical robots, under changing conditions, or across a wider range of tasks. The reported 18.56-second average applies to the study’s direct setup; it should not be treated as a universal latency figure for every configuration or future system.
The comparisons also have limits. Runway’s speed and cost figures came from its own test on one of its robots and three tasks. The available account does not establish whether the two models were evaluated under conditions that generalize across hardware, environments, or workloads. OpenAI has not publicly explained how the new organizational structure will affect staffing, research priorities, hardware development, or its humanoid-robot plans.
It remains unresolved whether general-purpose models can meet the speed and reliability requirements for real-world control, whether specialist models will remain necessary for fast movement, or how the two approaches might be combined. OpenAI’s planned public demonstration has been targeted for 2027, but the supplied reporting does not confirm a date or specify what the demonstration will show.
The 2027 Robot Demonstration
The clearest public milestone in the reporting is OpenAI’s stated target of a humanoid-robot demonstration in 2027. Until then, further details about the team’s placement, research plans, and hardware program would help show how the reorganization translates into work on robots. OpenAI had not posted a statement about the change on its news feed as of November 14.
For now, the relevant evidence will come from further benchmark results and tests on physical machines. Those results would need to report not just task success, but also response time, reliability, and cost under clearly described conditions. The current findings suggest different strengths for general-purpose and specialist models, but do not establish a winner.
Key Questions
What did OpenAI change?
OpenAI broke up its stand-alone robotics team. Fortune reported that robotics researchers moved into the main research organization under Mark Chen, while applied robotics and hardware teams moved into the Applied organization under Greg Brockman.
Why did OpenAI reorganize the team?
An OpenAI spokesperson told Fortune that general-purpose frontier models, including GPT-6 Astra, have shown they can be applied directly to robotics. The spokesperson said the company remains committed to robotics.
Did GPT-6 Astra outperform robot-specific models?
In one NVIDIA simulation study, an agentic GPT-6 Astra setup achieved a 74.8% success rate on RoboLab-120, compared with 51.0% for the best trained robot policy listed there. The result does not show that Astra is faster or more effective on physical robots; the study’s direct setup averaged 18.56 seconds per model call.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter robot-focused model released with weights and code under the Apache 2.0 license. Runway reported a 46.2% RoboLab-120 score and faster calls in its own comparison, but those results do not establish performance across all robots or tasks.
When might OpenAI show its humanoid robot?
CEO Sam Altman has targeted a first public demonstration in 2027. The reporting does not give a confirmed date or details about what the demonstration will include.
Source: ThorstenMeyerAI.com
