🔍 Read the full analysis: OpenAI's Robotics Team Splits As Runway Makes Praxis-1 Available 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 confirmed to Fortune that it has disbanded its stand-alone robotics team, moving researchers into its main research group and applied robotics and hardware teams into Applied. The company says general-purpose models can now be used directly for robot control; Runway, meanwhile, released the open-weight specialist model Praxis-1. Published tests suggest general models can perform well in simulation, while specialist models have a speed advantage, but the results do not establish which approach will work best in real-world robots.
OpenAI has broken up its stand-alone robotics team and moved its researchers into other parts of the company, Fortune reported November 13, with the company confirming the change when asked. The reorganization coincides with Runway’s release of Praxis-1, an open-weight model built specifically for robot action, sharpening a debate over whether robots need dedicated AI or can rely on general-purpose models.
According to Fortune, OpenAI’s robotics researchers now sit within its main research organization under Chief Research Officer Mark Chen. Its applied robotics and hardware teams have moved to the Applied organization, which reports to President Greg Brockman. Fortune said the company’s booth at the CoRL robotics conference in Austin remained listed on the floor plan but was empty when the outlet visited. OpenAI has not published its own announcement about the reorganization.
An OpenAI spokesperson told Fortune the change reflects a shift in the technology, not a retreat from robotics: the company had previously expected to build on video models such as Sora, but its general-purpose frontier models, including GPT-6 Astra, have since shown they can be applied directly to robot control. The spokesperson described robotics as part of OpenAI’s long-term vision and said the company remains committed to bringing general-purpose robots into the world.
Runway released Praxis-1 two days before Fortune’s report. The company describes it as a 27-billion-parameter “world action model” and released its weights and fine-tuning code under the Apache 2.0 license, which allows commercial use. On the RoboLab-120 simulation leaderboard cited in the source material, Praxis-1 scored 46.2%, ranking first among the listed world-action and open-source robot models and fifth overall. Runway reports that it takes about 0.55 to 0.6 seconds per call 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 Speed Gap in Robot Control
The competing releases point to a practical distinction between planning ability and response time. A general-purpose model may handle a wider range of tasks in a simulation, but a robot operating in the physical world also needs timely instructions. Slow calls can make a model difficult to use for actions that require rapid adjustments, even if it performs well on a benchmark.
In a study by researchers at NVIDIA’s Seattle Robotics Lab, GPT-6 Astra achieved a 74.8% success rate on RoboLab-120 in an agentic setup, compared with 51.0% for the best trained robot policy on the leaderboard cited in the report. The study’s direct setup averaged 18.56 seconds per model call, however, and paused the simulation while the model was thinking. That makes the result evidence of performance in a particular simulation, not proof that Astra can control a real robot at a useful speed.
Runway’s comparison points to the same trade-off from the other direction. In a company-run test on one of its robots, Runway 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 Runway’s own results and apply to its test setup; they do not establish how the models compare across other robots, tasks, or deployment conditions. The broader question is whether systems will combine general models for flexible reasoning with specialist models for fast movement, rather than choosing only one approach.
Products Worth Considering
LG 27US500-W 27-Inch 4K UHD IPS Monitor
Vari Ergo Electric Standing Desk, 54×26, Light Wood/White
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
As an affiliate, we earn on qualifying purchases.
OpenAI’s Shift Toward General Models
The reorganization marks the second time OpenAI has ended a stand-alone robotics team. The company wound down an earlier team in 2020. It rebuilt robotics work 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’s explanation for the current move is that its general models have become more applicable to robotics than the company expected earlier in the year. The approach has a different premise from a dedicated policy such as Praxis-1: rather than train a model specifically to issue robot actions, the general model is given images, information about the robot, and a control interface, then asked to choose movements.
OpenAI has said it is developing its own humanoid robot, and Chief Executive Sam Altman has targeted a public demonstration in 2027. Fortune also described pressure on the company to control costs and focus on core products, but that is background reported by the outlet; OpenAI’s stated explanation for the robotics reorganization was the progress of its general-purpose models.
“The two approaches are complementary.”
— Runway
Real-World Performance Remains Untested
The available results do not settle whether general-purpose or specialist models are better for deployed robots. RoboLab-120 is a simulation benchmark, and the cited GPT-6 Astra study tested specific configurations on tabletop tasks for a single simulated robot arm. The reported success rates do not show how either model performs across different hardware, settings, or extended real-world use.
Runway’s speed and cost comparison is based on one company’s test on one of its robots, completing three tasks. The source material does not provide enough detail to determine whether the same difference would hold for other tasks or systems. It also does not establish the degree to which Runway’s benchmark results are independently replicated.
OpenAI has not published a statement of its own on the reorganization, and the company’s internal reporting changes and plans for robotics remain described through Fortune’s account. The source material also does not specify when GPT-6 Astra might be used in a commercial robot, or how OpenAI intends to test it outside simulation.
Testing Models Beyond Simulation
The next evidence to watch for is real-world testing that reports performance, response time, reliability, and cost across comparable robot tasks. Further details from OpenAI about how its reorganized teams will develop or deploy robotics systems could clarify whether the company is pursuing a general-model-only approach or expects to combine general and specialist components.
Runway’s release gives researchers and companies access to Praxis-1’s weights and fine-tuning code under the stated Apache 2.0 license. That could allow outside users to test or adapt the model, though the source material does not identify a timetable for independent evaluations. OpenAI’s stated target of a public humanoid-robot demonstration in 2027 is another future milestone; no more specific date or demonstration details are provided.
Key Questions
What did OpenAI change?
According to Fortune, OpenAI ended its stand-alone robotics team. Its researchers moved into the main research organization 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 designed for robot action. Runway released its weights and fine-tuning code under the Apache 2.0 license, which permits commercial use.
Does the benchmark show GPT-6 Astra is better than Praxis-1?
Not by itself. In the cited simulation study, GPT-6 Astra reached a 74.8% success rate in an agentic setup, while Praxis-1 scored 46.2% on the RoboLab-120 leaderboard. The figures come from different model approaches and reported setups; they do not establish which model performs better on real robots.
Why did OpenAI reorganize its robotics work?
An OpenAI spokesperson told Fortune that general-purpose models, including GPT-6 Astra, had shown they could be applied directly to robotics. The company framed the move as a result of that progress and said robotics remains part of its long-term vision.
When might OpenAI show its humanoid robot?
The source material says CEO Sam Altman has targeted a public demonstration in 2027. It gives no specific date or details about what the demonstration would include.
Source: ThorstenMeyerAI.com
