🔍 Read the full analysis: OpenAI Splits Its Robotics Team As Runway Opens Praxis-1: Do Robots Need Their Own AI? on ThorstenMeyerAI.com
Get movie nights 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 confirmed to Fortune that it has reorganized its robotics teams, moving researchers into its main research organization and applied robotics and hardware teams into Applied. Runway released Praxis-1, an open-weight robot model, as published results show general models can perform well in simulation while specialist models act much faster.
OpenAI has disbanded its stand-alone robotics team and moved its researchers and applied robotics staff into other parts of the company, Fortune reported November 13, with OpenAI confirming the reorganization to the outlet. The shift came as Runway released Praxis-1, an open-weight model built for robot control, sharpening a practical question for the field: whether robots can use general-purpose AI models or need faster, specialized systems.
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 under President Greg Brockman. The change was recent enough that the company’s listed booth at the CoRL robotics conference in Austin was empty when Fortune visited. OpenAI has not published a separate announcement about the reorganization.
OpenAI’s spokesperson described the move to Fortune as a sign of continued commitment, not a retreat. The outlet quoted the company 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 spokesperson said the company had initially expected to build on video models such as Sora, but has since found that frontier models, including GPT-6 Astra, can be applied directly to robotics.
Runway’s Praxis-1 is a 27-billion-parameter “world action model,” released with weights and fine-tuning code under the Apache 2.0 license. In the RoboLab-120 benchmark figures cited by the source, Praxis-1 scored 46.2%, the highest among world action models and open-source robot models on the leaderboard, and fifth overall. Its reported calls take about 0.55 to 0.6 seconds to produce 1.6 seconds of motion.
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 competing approaches point to different strengths rather than a settled winner. In an NVIDIA researchers’ simulation study, an agent using GPT-6 Astra achieved a 74.8% success rate on RoboLab-120, above the 51.0% reported for the leaderboard’s best trained robot policy. But in the study’s direct setup, each model call averaged 18.56 seconds, and the simulation paused while the model processed its response. That delay matters when a robot needs to react to movement or changing conditions.
Runway reported a separate 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 results come from Runway’s own test, not an independent evaluation, and should not be treated as a general measure of performance across robots or tasks. Still, they illustrate why task success alone may not determine which model is useful in physical settings.
The organizational change also matters for how OpenAI develops robotics. Folding specialists into broader research and applied groups may make general models a more direct part of the company’s robotics work. It does not establish that OpenAI has stopped building robots or that its general models are ready for real-world deployment.
Popular right now
Hisense 55-Inch E6SR Series Hi-QLED 4K UHD Smart Roku TV (55E6SR, 2026 Model)
eufy E25 Robot Vacuum and Mop with HydroJet Roller Mop, 20,000 Pa Suction
KitchenAid Artisan 5-Qt Tilt-Head Stand Mixer, KSM150PS, Almond Cream
As an affiliate, we earn on qualifying purchases.
How OpenAI Reached This Reorganization
Fortune’s November 13 report says OpenAI’s team structure changed after researchers reconsidered how to connect AI models to robots. At the start of the year, they expected to build on models such as Sora, which generates video and represents how physical scenes behave. The company’s spokesperson later pointed to the ability of general-purpose frontier models to take on robotics tasks directly as the reason for moving the work into other groups.
This is the second time OpenAI has wound down a stand-alone robotics team. The company closed an earlier team in 2020. It rebuilt robotics research 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. OpenAI has also said it is developing a humanoid robot, with CEO Sam Altman targeting a public demonstration in 2027, according to the source material.
The evidence being discussed is not all from the same kind of test. NVIDIA’s work evaluated GPT-6 Astra in a simulation of 120 tabletop tasks involving one robot arm. Runway’s speed and cost comparison used one of its own robots and three tasks. Those results offer useful indicators, but they do not establish how either approach performs across different hardware, environments or safety requirements.
What the Results Do Not Establish
The reported benchmarks do not resolve whether general-purpose models can control robots reliably in everyday physical settings. RoboLab-120 is a simulation, while the Runway comparison covered one robot and three tasks. The tests use different setups, so their headline numbers are not a direct, independent comparison across the same full evaluation.
It is also unclear how OpenAI’s reorganization will affect staffing, research priorities or the development schedule for its own humanoid robot. The company has not posted a statement of its own about the change, and the timeline for any public demonstration remains based on the target attributed to Altman. The source material gives no independent validation of Runway’s reported speed and cost figures.
The 2027 Robot Demonstration Target
The next public milestone identified in the reporting is OpenAI’s targeted 2027 demonstration of its humanoid robot. The company has not provided further details in the source material about what that demonstration will show or whether the reorganization changes its schedule.
Further comparisons will need to test the models on the same hardware and tasks, with response time, cost and task success measured together. For now, OpenAI’s move and Runway’s release show two different development strategies, but neither the organizational change nor the cited benchmark results settle which approach will prove more useful outside controlled tests.
Key Questions
What happened to OpenAI’s robotics team?
Fortune reported that OpenAI disbanded its stand-alone team. Its robotics researchers moved under Chief Research Officer Mark Chen, while applied robotics and hardware teams moved into the Applied organization under Greg Brockman.
Why did OpenAI reorganize the team?
OpenAI’s spokesperson told Fortune that general-purpose models, including GPT-6 Astra, had shown they could be applied directly to robotics. The spokesperson presented the change as a commitment to the work, not a retreat.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter robot model released with downloadable weights and fine-tuning code under the Apache 2.0 license. Runway describes it as a “world action model.”
Does GPT-6 Astra outperform Praxis-1?
The reported figures do not provide a single, independent head-to-head answer. GPT-6 Astra scored higher in a cited simulation benchmark, while Runway reported that Praxis-1 completed three tasks on one of its robots faster and at lower cost.
When might OpenAI show its humanoid robot?
The source material says CEO Sam Altman has targeted a public demonstration in 2027. OpenAI has not provided further schedule details in the cited reporting.
Source: ThorstenMeyerAI.com
