🔍 Read the full analysis: Robot AI In The Spotlight After OpenAI Splits Its Team And Runway Opens Praxis-1 on ThorstenMeyerAI.com
Get smart everyday buys 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 moved its robotics researchers and applied robotics teams into other parts of the company, saying general-purpose models can now be applied directly to robotics. Runway has released the open-weight Praxis-1, a specialist model that scored lower than GPT-6 Astra on a cited simulation benchmark but acted much faster in the reported tests.
OpenAI has broken up its stand-alone robotics team and moved researchers into its main research group, while Runway has released Praxis-1, an open-weight model built specifically for robot control. The developments, reported around the CoRL robotics conference in Austin, sharpen a question for the field: whether general-purpose AI can direct robots effectively, or specialist systems remain necessary for fast, real-time action.
Fortune reported on November 13 that OpenAI confirmed the reorganization after its booth at CoRL was found empty despite appearing on the conference floor plan. The robotics researchers now report to Chief Research Officer Mark Chen; applied robotics and hardware teams have moved to the Applied organization, under President Greg Brockman. OpenAI has not published a separate statement about the change, according to the source material.
An OpenAI spokesperson told Fortune the move reflects the company’s view that general-purpose frontier models, including GPT-6 Astra, can be used directly for robotics. The spokesperson characterized robotics as part of OpenAI’s long-term vision, rather than saying the company was abandoning the work. The company has previously said it is developing a humanoid robot, with CEO Sam Altman targeting a public demonstration in 2027.
Runway released Praxis-1 two days before the November 14 report. The company describes it as a 27-billion-parameter world action model and made its weights and fine-tuning code available under the Apache 2.0 license. In the cited RoboLab-120 results, Praxis-1 scored 46.2%, the highest result among the world action models and open-source robot models listed, and fifth overall. Runway’s reported timing was about 0.55 to 0.6 seconds per call for 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
Speed Versus Task Success
The results point to a practical divide between what a model can accomplish and how quickly it can control a machine. In a simulation study by researchers at NVIDIA’s Seattle Robotics Lab, GPT-6 Astra reached a 74.8% success rate on RoboLab-120 when used in an agentic setup. The study compared that result with 51.0% for the best trained robot policy on the leaderboard at the time. But in a direct setup, each Astra call took an average of 18.56 seconds, with the simulation paused while the model processed its response.
For robots interacting with changing surroundings, long waits can limit usefulness even when a model performs well on a set of tasks. Runway’s specialist model had a lower benchmark score, but its reported calls were much faster. These figures come from different tests and setups, so they do not establish a definitive head-to-head result. They do, however, show why model accuracy alone may not settle which approach works best outside a simulation.
Runway said the approaches could be complementary: a general model could reason through a task, while a faster specialist handles immediate movement. In Runway’s own test on one of its robots, the company 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 comparisons are company-reported and relate to that specific test, not a general measure of robot performance or operating cost.
Products Worth Considering
HP Sprocket Studio Plus Wireless 4×6 Photo Printer
Brother SP1 Dye-Sublimation Printer
Dell UltraSharp U2424H 24-Inch FHD Monitor
As an affiliate, we earn on qualifying purchases.
OpenAI’s Robotics Team History
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 pointed to limited robot data and the prospect of pursuing artificial general intelligence more efficiently through other approaches. OpenAI 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 the partnership.
The latest reorganization follows a change in OpenAI’s stated technical rationale. At the start of 2026, its researchers expected to build on video models such as Sora, according to the account provided to Fortune. The spokesperson later said newer general-purpose models, including GPT-6 Astra, had shown they could be applied to robotics directly. Fortune also cited pressure on OpenAI to control costs and focus on core products as business background, but the reason OpenAI gave for the team move was the progress of its general models.
The benchmark evidence comes from an NVIDIA researchers’ study published October 14, 2026, as a CoRL 2026 late-breaking result. RoboLab-120 is a simulation of 120 tabletop tasks for a single robot arm. The study tested Astra as a robot policy using camera images, robot information and a control interface, without robot-specific training. Its agentic setup reached 74.8%; a direct setup reached 66.3%, and an orchestrated setup in which Astra reviewed movements from a trained policy reached 66.5%.
“Complementary.”
— Runway
Limits of the Robotics Comparisons
The cited figures do not establish which approach is better across real-world robots. The RoboLab-120 results are simulated, while Runway’s time and cost comparison came from its own test on one of its robots and involved three tasks. The available reporting does not provide enough detail to treat those results as a controlled, independent comparison across a broader range of machines or conditions.
It is also unclear how OpenAI will organize or resource robotics work after the reorganization, whether it will continue developing its own humanoid hardware, or how the move affects the previously stated target of a public demonstration in 2027. OpenAI had not posted a dedicated announcement on its news feed as of November 14, according to the source material. Further details about Praxis-1’s performance in independent tests were not provided.
Tests Beyond the Simulation
The next evidence to watch for is independent testing of Praxis-1 and general-purpose models on the same robots, tasks and timing conditions. Such comparisons would help show whether the simulation performance and company-reported speed figures translate across different hardware and real operating environments.
OpenAI’s next public milestone may be its stated plan for a humanoid robot demonstration in 2027, though the reorganization report does not confirm whether that schedule has changed. For now, the two companies’ moves show different organizational choices: OpenAI is integrating robotics researchers into broader groups, while Runway is making a dedicated model available under an open license.
Key Questions
What changed at OpenAI?
OpenAI moved its robotics researchers into its main research organization under Mark Chen and moved applied robotics and hardware teams to the Applied organization under Greg Brockman, according to Fortune.
What is Praxis-1?
Praxis-1 is Runway’s 27-billion-parameter world action model for robot control. Runway released its weights and fine-tuning code under the Apache 2.0 license.
Which model performed better on RoboLab-120?
In the cited simulation results, GPT-6 Astra scored 74.8% in an agentic setup, while Praxis-1 scored 46.2%. The figures come from different reported evaluations and are not a direct, controlled comparison.
Why might a specialist robot model still be useful?
The available figures suggest a possible speed advantage: Runway reported Praxis-1 taking about 0.55 to 0.6 seconds per call, while NVIDIA researchers reported an 18.56-second average per call for Astra in a direct setup. The tests differ, and the results do not prove that specialist models are faster in every setting.
Has OpenAI abandoned its robotics plans?
OpenAI’s spokesperson told Fortune that robotics remains part of the company’s long-term vision. The company has previously stated a target of a public humanoid robot demonstration in 2027; whether the reorganization changes that plan is not clear from the reporting.
Source: ThorstenMeyerAI.com
