🔍 Read the full analysis: OpenAI Splits Its Robotics Team While Runway Opens A New Chapter With Praxis-1 on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI has dissolved its stand-alone robotics team, moving researchers into its main research organization and applied robotics and hardware staff into its Applied organization. Two days before the change was reported, Runway released Praxis-1, an open-weight robot model; available tests suggest general models can perform more simulated tasks, while specialist models respond much faster.
OpenAI has broken up its stand-alone robotics team and moved its researchers and applied robotics staff into other parts of the company, Fortune reported on November 13, while Runway released Praxis-1, a dedicated robot model, two days earlier. The developments highlight different approaches to robot control: OpenAI says its general-purpose models can be applied directly, while Runway is offering a specialist model with downloadable weights.
According to Fortune, which said OpenAI confirmed the reorganization when asked, the robotics researchers now report to Chief Research Officer Mark Chen. The applied robotics and hardware teams moved to the Applied organization, which reports to President Greg Brockman. OpenAI had registered a booth at the CoRL robotics conference in Austin, but the booth was empty when Fortune visited.
An OpenAI spokesperson told Fortune that the move reflects a shift in how the company approaches robotics, not an abandonment of the field. The spokesperson said that general-purpose frontier models, including GPT-6 Astra, have shown they can be used for robot control directly. OpenAI has not published its own announcement about the reorganization, according to the source material.
Runway’s Praxis-1 is a 27-billion-parameter world action model released with weights and code under the Apache 2.0 license, which permits commercial use subject to the license terms. Runway says it can be fine-tuned. On the RoboLab-120 benchmark, Praxis-1 scored 46.2% success, ranking fifth overall and first among the world action and open-source robot models listed in the reported results.
A separate NVIDIA study tested GPT-6 Astra on the same simulation benchmark, which covers 120 tabletop tasks for a single robot arm. Its agentic setup reached 74.8% success, compared with 51.0% for the best trained robot policy on the leaderboard cited in the report. That result came with a substantial delay: in the study’s direct setup, each model call averaged 18.56 seconds, and the simulation paused while the model processed 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 results point to a practical distinction for robotics developers: task success is not the only measure of whether a model can control a physical robot. A system that performs well in a paused simulation may still be too slow for a machine that must react continuously to movement, contact, or changing surroundings.
Runway reported a separate test on one of its robots in which GPT-6 Astra took 15.6 times longer than Praxis-1 to finish three tasks and cost 51 times more per completed task. Those figures are Runway’s own results, not an independent comparison, and the source does not provide enough detail to assess how broadly they apply. Still, the comparison illustrates why specialist models may remain useful even if general models handle a wider range of tasks.
The organizational change could also affect how OpenAI develops robot systems. By placing robotics researchers alongside general-model teams, the company is betting that progress in its main models can carry over to physical control. Runway’s release offers developers a different option: a model built specifically for robot actions, with weights available for adaptation. Neither announcement establishes which approach will work best in real-world deployment.
Popular right now
Dyson V15 Detect Cordless Vacuum
De’Longhi Classic Matte Black Espresso Machine with Milk Frother
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Cypress Green
As an affiliate, we earn on qualifying purchases.
OpenAI’s Earlier Robotics Efforts
OpenAI has pursued robotics before. It wound down an earlier stand-alone team in 2020. Cofounder Wojciech Zaremba later cited limited robot data and a more efficient route to artificial general intelligence through other methods. The company 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 that partnership.
OpenAI’s current explanation differs from the earlier rationale: its spokesperson told Fortune that general-purpose models have become more directly useful for robotics. The company has also said it is developing a humanoid robot, and CEO Sam Altman has targeted a public demonstration in 2027. Fortune mentioned cost controls and product focus as wider business pressures, but OpenAI’s stated reason for the team change was its view of general models’ capabilities.
The evidence cited for that view includes an NVIDIA Seattle Robotics Lab study published October 14, 2026, as a CoRL 2026 late-breaking result. It tested Astra without robot-specific training, giving it camera images, information about the robot, and a control interface. Its results came from a simulation benchmark; they do not by themselves show how the system performs on physical robots.
What the Results Do Not Establish
OpenAI has not publicly detailed the reorganization or explained how robotics work will be staffed and funded across the two receiving organizations. The Fortune report says the company confirmed the change when asked, but no standalone OpenAI statement was available in the source material.
The benchmark results also leave important questions open. RoboLab-120 is a simulation, and the reported success rates do not establish how reliably either model operates on a physical robot across varied environments. The NVIDIA study’s agentic setup reached 74.8%, but its long model-call time limits what that score alone says about real-time control. Runway’s speed and cost comparison covers three tasks on one of its robots; independent testing and more details on the test conditions would be needed to judge how general the comparison is.
It is also unclear whether OpenAI will continue developing a robot of its own, how its reorganized teams will collaborate on that effort, or whether general models and specialist policies will be combined in a deployed system. The reported 2027 demonstration target is a goal, not confirmation of a scheduled public event.
The 2027 Robot Demonstration Target
The next public milestone identified in the source material is OpenAI CEO Sam Altman’s target of a humanoid robot demonstration in 2027. OpenAI has not provided details here about a date, venue, robot capabilities, or whether the reorganization changes that target.
For Praxis-1, the immediate next test is whether developers can reproduce or extend Runway’s reported results using its released weights and code. Further evaluations that compare models on the same physical robots, tasks, and timing conditions would help clarify the trade-offs between general capability, response speed, and cost. Until then, the available evidence is limited to reported benchmark and company-run results.
Key Questions
What did OpenAI change in its robotics organization?
Fortune reported that OpenAI ended its stand-alone robotics team. Robotics researchers now report to Chief Research Officer Mark Chen, while applied robotics and hardware staff moved into the Applied organization under President Greg Brockman.
Why did OpenAI reorganize the team?
An OpenAI spokesperson told Fortune that general-purpose models, including GPT-6 Astra, have shown they can be applied directly to robotics. The company framed the move as a change in approach, not a withdrawal from robotics.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter model designed for robot actions. Runway released its weights and code under the Apache 2.0 license and reported a 46.2% success rate on the RoboLab-120 benchmark.
Does GPT-6 Astra outperform Praxis-1?
They were reported at different scores and in distinct evaluations. Astra reached 74.8% in an agentic setup on the simulated RoboLab-120 benchmark, while Praxis-1 scored 46.2%; Astra’s reported direct calls were much slower. The figures do not establish which model is better for real-world robot use.
When might OpenAI show its humanoid robot?
CEO Sam Altman has targeted a public demonstration in 2027, according to the source material. No date or further event details were provided, so the timing remains a target rather than a confirmed schedule.
Source: ThorstenMeyerAI.com
