🔍 Read the full analysis: Runway's Praxis-1 Opens As OpenAI's Robotics Team Splits Up on ThorstenMeyerAI.com
Get hardware and tech essentials 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 moved its robotics researchers into its main research organization and its applied robotics and hardware teams into its Applied organization. The change comes as Runway releases Praxis-1, an open-weight robot model; available tests show general models can perform well in simulation, while specialist models act faster.
OpenAI has broken up its stand-alone robotics team, moving its researchers into the company’s main research organization and its applied robotics and hardware teams into its Applied organization, according to Fortune. The change coincides with Runway’s release of Praxis-1, an open-weight model designed for robot control, sharpening a debate over whether robotics needs specialist AI models or can rely on general-purpose systems.
Fortune reported on November 13 that OpenAI confirmed the reorganization after the publication asked about it. The robotics researchers now report to Chief Research Officer Mark Chen, while the applied robotics and hardware teams moved to the Applied organization, which reports to President Greg Brockman. OpenAI had a booth listed at the CoRL robotics conference in Austin, but Fortune found it empty.
An OpenAI spokesperson described the change as a sign of continuing commitment, not a retreat. The spokesperson told Fortune that robotics remains part of the company’s long-term vision and said general-purpose frontier models, including GPT-6 Astra, have shown they can be applied directly to robotics. OpenAI has not published its own announcement about the reorganization; its public news feed had no item on the change as of November 14, according to the source material.
Runway released Praxis-1 two days before the Fortune report. The company describes it as a 27-billion-parameter “world action model.” Its weights and fine-tuning code are available under the Apache 2.0 license, which permits commercial use subject to that license’s terms. On RoboLab-120, Praxis-1 scored 46.2%, placing it fifth overall and ahead of the other world action and open-source robot models listed in the reported comparison.
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
Why Robot Models Face a Speed Test
The reorganization and release highlight two different approaches to putting AI into machines. General-purpose models may handle a wider range of tasks without robot-specific training, while specialist models may be better suited to quick, repeated control. That distinction matters because a system that performs well in a paused simulation may still be too slow for a robot operating continuously in the physical world.
In an NVIDIA Seattle Robotics Lab simulation study, GPT-6 Astra’s agentic setup reached a 74.8% success rate on RoboLab-120, a benchmark of 120 tabletop tasks for a single robot arm. The researchers reported that its average call took 18.56 seconds in the direct setup, with the simulation pausing while the model reasoned. By comparison, Praxis-1 takes about 0.55 to 0.6 seconds per call for 1.6 seconds of motion, according to the figures in the source material.
Those results do not establish that either system is better for real-world deployment. They do show why model quality alone is not the only measure: response time, cost, reliability and performance outside simulation also matter. Runway says general and specialist approaches can work together, rather than treating the choice as either-or.
Products Worth Considering
Bose QuietComfort Wireless Headphones – Twilight Blue (Limited Edition)
LG UltraGear evo 27-inch 5K Mini-LED Gaming Monitor
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
As an affiliate, we earn on qualifying purchases.
OpenAI’s Earlier Robotics Efforts
The current change is OpenAI’s second closure of a stand-alone robotics team. The company wound down its first robotics group in 2020. Cofounder Wojciech Zaremba later pointed to limited robot data and a more efficient route to artificial general intelligence through other methods. OpenAI rebuilt a robotics team 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 latest move is a shift in its technical assessment. Earlier in 2026, researchers expected to build robotics work on video models such as Sora. The spokesperson told Fortune that newer general-purpose models, including GPT-6 Astra, changed that view by showing they could be applied to robot control directly. Fortune also described cost-cutting and a focus on core products as business context, but that was not the reason OpenAI gave for the team changes.
The reported GPT-6 Astra results come from a simulation study published October 14, 2026, by NVIDIA researchers and labelled a CoRL 2026 late-breaking result. In addition to the 74.8% agentic-policy score, the study reported 66.3% for a direct policy and 66.5% for an orchestrated policy, in which Astra reviews movements proposed by a trained robot model. These are benchmark results, not evidence of equivalent performance in homes, factories or other live settings.
“The two approaches are complementary.”
— Runway
What Simulation Results Cannot Settle
It remains unclear how the reorganization will affect OpenAI’s robotics projects, staffing or development schedule. OpenAI has not issued a public statement with further details, and the report does not specify whether the company plans to keep developing a humanoid robot or how the teams will coordinate after the move.
The benchmark figures also leave practical questions unanswered. RoboLab-120 is a simulation of tabletop tasks, and the cited results do not establish how either model performs across different robots, changing environments or sustained real-world use. The source material gives a same-robot comparison in which GPT-6 Astra took 15.6 times longer than Praxis-1 to complete three tasks and cost 51 times more per completed task, based on Runway’s own test. Those figures are company-reported and apply to that test; test methodology and broader comparability are not established here.
It is also not clear whether OpenAI’s reported team move reflects a permanent operating structure or a change that may evolve. The company’s spokesperson characterized the decision as an acceleration of its robotics work, but no timeline or detailed plan accompanied the account.
Signals to Watch From Both Companies
Further information from OpenAI may clarify how its research and applied teams will pursue robotics, and whether it will publish details about the reorganization. The source material says CEO Sam Altman has targeted a first public demonstration of OpenAI’s humanoid robot in 2027; that is a target, not confirmation that a demonstration will take place on a set date.
For Runway, the next evidence will be how Praxis-1 performs beyond the reported benchmark and company test, including whether outside developers can reproduce its results and adapt the model to different robots. More comparisons using shared tasks, hardware and reporting methods could help establish where fast specialist policies and more general models each fit.
For now, the confirmed developments are organizational and technical: OpenAI has redistributed its robotics teams, and Runway has released Praxis-1 with open weights. Neither announcement resolves whether general-purpose AI can match the speed and reliability needed for practical robot control.
Key Questions
What did OpenAI change in its robotics organization?
According to Fortune, OpenAI moved its robotics researchers into the main research organization under Chief Research Officer Mark Chen. Its applied robotics and hardware teams moved to the Applied organization under President Greg Brockman.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter robot model that Runway calls a world action model. Its weights and fine-tuning code are released under the Apache 2.0 license for use subject to that license’s terms.
Did GPT-6 Astra outperform Praxis-1?
It depends on the measure. In the cited RoboLab-120 simulation, an agentic GPT-6 Astra setup scored 74.8%, compared with Praxis-1’s reported 46.2%. The reported call times favor Praxis-1, and the results do not establish which model performs better in real-world use.
Has OpenAI left robotics?
OpenAI’s spokesperson told Fortune that robotics remains part of the company’s long-term vision. The reported change is a move away from a stand-alone robotics team, not a confirmed withdrawal from robotics.
When might OpenAI show its humanoid robot?
The source material says CEO Sam Altman has targeted a first public demonstration in 2027. That is a stated target; no specific date or confirmation of a scheduled demonstration is provided.
Source: ThorstenMeyerAI.com
