🔍 Read the full analysis: Runway Opens Praxis-1 While OpenAI Reorganizes Its Robotics Team 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
Runway released Praxis-1, an open-weight robot model, two days before Fortune reported that OpenAI had dispersed its stand-alone robotics team across other company groups. The available benchmark results point to a trade-off: general models performed better on one simulation benchmark, while Praxis-1 acted faster; real-world performance and the companies’ longer-term plans remain uncertain.
Runway released Praxis-1, an open-weight model for robot control, two days before Fortune reported that OpenAI had broken up its stand-alone robotics team and moved its researchers into broader company groups. The developments highlight competing approaches to robot intelligence, but the available results show a trade-off rather than a clear winner: a general-purpose model scored higher in one simulation test, while Runway’s specialist model responded faster.
Fortune reported on November 13 that OpenAI confirmed the reorganization after the outlet found its booth empty at the CoRL robotics conference in Austin. According to Fortune, robotics researchers now report to Chief Research Officer Mark Chen, while applied robotics and hardware teams moved to the Applied organization under President Greg Brockman. OpenAI had registered a booth for the conference, but it was unstaffed when Fortune visited. OpenAI has not published its own announcement about the change.
An OpenAI spokesperson told Fortune the move reflects the company’s view that its general-purpose frontier models, including GPT-6 Astra, can be applied directly to robotics. 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. That is the company’s stated explanation; Fortune also described pressure on OpenAI to control costs and focus on core products, but did not report that as the reason the company gave for the reorganization.
Runway’s Praxis-1 is a 27-billion-parameter “world action model.” Runway released its weights and fine-tuning code under the Apache 2.0 license, which permits commercial use under that license’s terms. On the RoboLab-120 leaderboard cited in the source material, Praxis-1 scored 46.2%, the highest result among world action models and open-source robot models, and fifth overall. Runway reports that its model 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
Two Routes to Robot Control
The developments matter because they put different strategies for robot intelligence in direct view. OpenAI’s reorganization suggests it is placing robotics work closer to the general models it says can already control robots. Runway, by contrast, is releasing a model designed specifically for action, with weights and code available for others to inspect, adapt and use commercially.
The benchmark figures do not settle which approach is better. In a study by NVIDIA researchers, an agent using GPT-6 Astra achieved a 74.8% success rate on RoboLab-120 in an agentic setup, compared with 46.2% for Praxis-1 on the cited leaderboard. But the study’s direct GPT-6 Astra setup averaged 18.56 seconds per model call, and the simulation paused while the model generated a response. Runway reports much shorter calls for Praxis-1. These measurements come from different reported evaluations and should not be treated as a controlled, head-to-head comparison.
Speed and task success have practical consequences for robots that must respond to changing surroundings. A model that succeeds often in simulation may still be too slow for a particular task, while a faster model may not handle as many tasks reliably. The available evidence points to complementary roles, as Runway itself has characterized the approaches, rather than proving that general models or specialist models will replace the other.
Products Worth Considering
Bose QuietComfort Headphones (2nd Gen), Wireless Over-Ear Noise-Cancelling Headphones, Black
Jackery Solar Generator 1000 v2 with 100W Solar Panel
Jackery Explorer 2000 v2 Portable Power Station with Two 200W Solar Panels
As an affiliate, we earn on qualifying purchases.
OpenAI’s Robotics Team History
The move is the second time OpenAI has closed a stand-alone robotics team. It wound down its earlier team in 2020; co-founder Wojciech Zaremba later cited limited robot data and a more efficient path 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 change is that general-purpose models have become more directly useful for robot control than the company expected earlier in 2026, when researchers anticipated building on video models such as Sora. Fortune also reported that OpenAI is developing a humanoid robot and that CEO Sam Altman has targeted a public demonstration in 2027. That target is a reported plan, not a confirmed demonstration date.
The performance evidence cited alongside the news comes from an October 14, 2026 study by researchers at NVIDIA’s Seattle Robotics Lab, presented as a CoRL 2026 late-breaking result. The study used RoboLab-120, a simulation of 120 tabletop tasks for a single robot arm. It tested GPT-6 Astra in several configurations, including a direct policy, an agentic setup with persistent history and extra tools, and an arrangement in which Astra reviewed proposals from a trained robot model.
What the Benchmarks Do Not Show
The results do not establish how either model performs across real-world robots, varied environments or extended use. RoboLab-120 is a simulation of tabletop tasks for one robot arm, and the figures cited come from separate reported evaluations. The source material does not provide enough detail to treat Runway’s results and the NVIDIA study as a fully matched comparison of hardware, task conditions and evaluation procedures.
It is also not clear whether OpenAI will continue developing robot-specific systems within its new organizational structure, how the research and applied teams will coordinate, or whether the company will release further technical details. OpenAI’s explanation was provided to Fortune; the company had not posted its own account of the change as of November 14, 2026. The reported 2027 humanoid demonstration remains a target, with no confirmed date or scope in the supplied material.
Evidence From Robots in Practice
The next meaningful test will be whether either approach can move beyond simulation and perform reliably on physical robots. For Praxis-1, the release of weights and fine-tuning code gives outside researchers and companies a way to test and adapt the model, though the source material does not describe independent real-world evaluations yet.
For OpenAI, developments to watch include any company statement clarifying how robotics work will operate within the research and Applied organizations, further technical results, and progress toward the humanoid demonstration target reported by Fortune. Until those details emerge, the confirmed news is an organizational change at OpenAI and a separate open-model release from Runway—not a settled verdict on which kind of model will control future robots.
Key Questions
What is Praxis-1?
Praxis-1 is Runway’s 27-billion-parameter model for robot action. Runway released its weights and fine-tuning code under the Apache 2.0 license.
What changed at OpenAI?
According to Fortune, OpenAI moved robotics researchers under Chief Research Officer Mark Chen and moved applied robotics and hardware teams into the Applied organization under President Greg Brockman. OpenAI confirmed the change to Fortune but had not issued its own public announcement as of November 14, 2026.
Did GPT-6 Astra outperform Praxis-1?
GPT-6 Astra reached a 74.8% success rate in an agentic configuration in the NVIDIA researchers’ RoboLab-120 study. Runway reported a 46.2% Praxis-1 result on the cited leaderboard. Because these are not presented as a matched head-to-head test, the figures do not establish an overall winner.
Why might a specialist robot model still be useful?
The reported tests suggest a speed trade-off. The NVIDIA study said GPT-6 Astra’s direct setup averaged 18.56 seconds per call, while Runway reported about 0.55 to 0.6 seconds per Praxis-1 call. The measurements come from different evaluations, and real-world performance remains unclear.
Source: ThorstenMeyerAI.com
