🔍 Read the full analysis: OpenAI Changes Its Robotics Team As Runway Opens Praxis-1 To The Public on ThorstenMeyerAI.com
Get the latest gadgets 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, moving its researchers into the main research group and its applied robotics and hardware teams into Applied, Fortune reported. Two days before the report, Runway released Praxis-1, an open-weight robot model. Available results show a trade-off: a general model performed better in one simulation benchmark, while specialist models were much faster.
OpenAI has broken up its stand-alone robotics team and moved its researchers into other parts of the company, Fortune reported November 13, while Runway opened access to Praxis-1, a dedicated robot-control model, two days earlier. The developments put competing approaches in focus: using general-purpose AI models to control robots, or building specialized models designed for faster physical actions.
Fortune said OpenAI confirmed the reorganization after the outlet asked about it. Robotics researchers now report to Chief Research Officer Mark Chen, while applied robotics and hardware teams have moved to the Applied organization under President Greg Brockman. OpenAI had a booth listed at the CoRL robotics conference in Austin, but it was empty when Fortune visited. OpenAI had not published its own announcement as of November 14.
The company’s spokesperson described the change as a sign of commitment rather than a retreat. According to Fortune, OpenAI said its frontier models, including GPT-6 Astra, had shown they could be applied directly to robotics. Earlier in the year, researchers had expected to build on video models such as Sora. OpenAI has also said it is developing a humanoid robot, with CEO Sam Altman targeting a public demonstration in 2027.
Runway released Praxis-1 as a 27-billion-parameter world action model, with weights and fine-tuning code under the Apache 2.0 license. In the reported RoboLab-120 results, it scored 46.2%, the highest score among the world action models and open-source robot models on the leaderboard, and fifth overall. Runway’s reported inference time was about 0.55 to 0.6 seconds per call for 1.6 seconds of motion. These figures come from separate evaluations and should not be treated as a direct comparison with every GPT-6 Astra result.
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 Reach and Speed
The competing announcements matter because a robot model must do more than choose a plausible action: it must produce instructions quickly enough for a physical system to respond. Simulation success and real-time control are different measures. A model that handles a wider range of tasks may still be too slow for many actions, while a faster specialized model may perform less broadly.
In a simulation study by NVIDIA researchers, an agent using GPT-6 Astra achieved 74.8% success on RoboLab-120, compared with 51.0% for the best trained robot policy on that leaderboard. But the study’s direct setup averaged 18.56 seconds per model call, and the simulation paused while the model processed its response. That result does not establish that Astra can control a physical robot at a comparable speed.
Runway reported a different test on one of its own robots: GPT-6 Astra took 15.6 times longer than Praxis-1 to complete three tasks and cost 51 times more per completed task. That comparison is Runway’s own test, not an independent benchmark, and the supplied account does not give enough detail to generalize it across robots or tasks. Taken together, the results point to practical trade-offs rather than a clear winner.
Products Worth Considering
Samsung Odyssey G55C 32-Inch QHD 1000R Curved Gaming Monitor
DJI Mini 5 Pro Fly More Combo Plus with DJI RC 2
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
As an affiliate, we earn on qualifying purchases.
OpenAI’s Robotics Reorganization
This is the second time OpenAI has closed a stand-alone robotics team. The company wound down its first team in 2020. It rebuilt a robotics group in 2024 and partnered with humanoid-maker Figure; Figure said in February 2025 that it would develop its own robot AI, ending that partnership.
Fortune reported that OpenAI’s current explanation is a shift in how the company expects robotics work to proceed: general-purpose frontier models may be usable directly, rather than requiring a separate robotics organization centered on adapting video models. The report also described pressure on OpenAI to control costs and focus on core products, but the company’s stated reason was the capabilities of its general models; the cost pressure was background, not a reason OpenAI gave for this reorganization.
The NVIDIA study, published October 14, 2026, tested Astra as a native robot policy in simulation, using camera images, robot information and a control interface without robot-specific training. It compared direct API use, an agentic setup with persistent history and extra tools, and a setup in which Astra reviewed a trained policy’s proposed movements. The 74.8% result came from the agentic setup, not every configuration.
What the Results Do Not Show
OpenAI has not published a direct statement detailing the reorganization, and the available reporting does not specify how many people moved, whether roles were cut, or what projects will continue under the new structure. It is also unclear how the new reporting lines will affect the company’s stated plan for a humanoid robot demonstration in 2027.
The benchmark figures do not establish how either approach performs across real-world robots and conditions. The NVIDIA result was produced in simulation, with the simulation pausing during model processing. Runway’s speed and cost comparison was conducted on one of its robots for three tasks, according to the supplied report; its methodology and broader applicability are not established there. Neither result alone settles whether general or specialist models are better for deployment.
Evidence to Watch in 2027
The next useful evidence will be independent tests on physical robots that report success, response time and operating cost across the same tasks and hardware. Further results could clarify whether general models can meet real-time requirements, whether specialist models can handle a wider range of tasks, and how the two might be combined.
Readers can also watch for details from OpenAI about its reorganized teams and the company’s stated target of a public humanoid-robot demonstration in 2027. For Praxis-1, the release of downloadable weights and fine-tuning code gives outside researchers and developers a route to test the model, but public availability is not itself evidence of performance in commercial settings.
Key Questions
What did OpenAI change?
Fortune reported that OpenAI ended its stand-alone robotics team. Robotics researchers moved to the main research group under Mark Chen, while applied robotics and hardware teams moved to the Applied organization under Greg Brockman.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter model for robot actions. Runway released its weights and fine-tuning code under the Apache 2.0 license, allowing public use under that license’s terms.
Did GPT-6 Astra outperform Praxis-1?
There is no single direct result establishing that. Astra scored 74.8% in an NVIDIA simulation study’s agentic setup, while Praxis-1 scored 46.2% on the reported RoboLab-120 leaderboard. The figures reflect different models and evaluation setups; Runway separately reported a speed and cost comparison on its own robot.
Why might a specialist model still be useful?
The reported results suggest that specialist models can respond faster. Runway reported Praxis-1 took about 0.55 to 0.6 seconds per call for 1.6 seconds of motion, while the NVIDIA study’s direct Astra setup averaged 18.56 seconds per call. These measurements come from different tests and are not a controlled head-to-head comparison.
What remains unknown about OpenAI’s plans?
OpenAI has not publicly detailed staffing, project changes or how the new organization will affect its humanoid-robot work. The company has said it is developing a humanoid robot and has targeted a public demonstration in 2027, but the supplied reporting does not confirm further details.
Source: ThorstenMeyerAI.com
