🔍 Read the full analysis: Runway's Praxis-1 And OpenAI's Team Split Bring Robot AI To The Fore on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
OpenAI confirmed it has dispersed its stand-alone robotics team across its research and applied organizations, saying its general-purpose models can now be applied directly to robot control. Runway has released Praxis-1, an open-weight specialist model; available tests show a trade-off between general models’ simulation results and specialist models’ speed. How the approaches will be combined in deployed robots remains unresolved.
OpenAI has broken up its stand-alone robotics team and moved researchers into its main research organization, while shifting applied robotics and hardware staff to its Applied organization, according to Fortune. The change comes as Runway releases Praxis-1, an open-weight robot model, and as published tests suggest general-purpose AI can handle a wider range of simulated tasks while specialist systems respond faster.
Fortune reported on November 13 that OpenAI confirmed the reorganization when asked. Its robotics researchers now report to Chief Research Officer Mark Chen; applied robotics and hardware teams moved to the Applied organization, led by President Greg Brockman. The company had registered a booth at the CoRL robotics conference in Austin, but Fortune found it empty. OpenAI has not published its own announcement about the change.
OpenAI’s spokesperson told Fortune the change reflects a shift in how the company believes robotics should be developed. Researchers had expected earlier in the year to build on video models such as Sora, but the spokesperson said general-purpose frontier models, including GPT-6 Astra, have since shown they 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.
Runway released Praxis-1 two days before the Fortune report. The model has 27 billion parameters and is described as a “world action model.” Runway made its weights and fine-tuning code available under the Apache 2.0 license. On RoboLab-120, it scored 46.2%, the highest result among the world action models and open-source robot models listed in the reported comparison, and fifth overall.
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 Simulation Success
The results point to a practical trade-off, not a settled winner. In an NVIDIA researchers’ simulation study, GPT-6 Astra achieved a 74.8% success rate on RoboLab-120 when used in an agentic setup. That exceeded the reported 51.0% result for the best trained robot policy on the benchmark’s leaderboard. But the study’s direct setup averaged 18.56 seconds per model call, and the simulation paused while Astra generated its response.
Runway’s reported timing for Praxis-1 was about 0.55 to 0.6 seconds per call, producing 1.6 seconds of motion. In a separate 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 come from Runway’s own test, not an independent evaluation, and the available source does not provide enough detail to establish how broadly they apply.
For robotics developers, the distinction matters because a model must do more than choose a successful action eventually: it must respond at a speed and cost suitable for the task and hardware. General-purpose models may offer flexibility across tasks, while specialized models may be better suited to rapid control. Runway called the approaches “complementary,” leaving open the possibility that general reasoning and fast robot policies could work together.
Popular right now
Ninja CREAMi 7-in-1 Ice Cream Maker, NC301
TCL 65-Inch 4K UHD Smart LED TV (65S425)
Dyson V8 Cyclone Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
OpenAI’s Shifting Robotics Strategy
This is the second time OpenAI has closed a stand-alone robotics team. The company wound down its first team in 2020. It rebuilt a robotics group 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 has said it is developing a humanoid robot, and CEO Sam Altman has targeted a public demonstration in 2027. That is a stated target, not confirmation that a demonstration or product will be ready by then. Fortune also reported that OpenAI faces pressure to control costs and focus on core products, but the company’s stated explanation for the team change was its view that general models can now be applied to robotics directly.
The RoboLab-120 result cited in coverage came from NVIDIA researchers’ October 14, 2026 study, described as a CoRL 2026 late-breaking result. The benchmark uses simulation for 120 tabletop tasks performed by one robot arm. The reported score depended on the setup: Astra scored 66.3% as a direct policy, 74.8% in an agentic setup with persistent history and extra tools, and 66.5% when it reviewed another model’s proposed movements. These are simulation findings, not evidence of performance across physical robots or real-world settings.
“The two approaches are complementary.”
— Runway
Questions on Real-World Robot Performance
The reported numbers do not settle whether general-purpose models or specialist robot models will perform better in physical deployment. NVIDIA’s result is from simulation**, while Runway’s timing and cost comparison comes from a company test on one of its robots. The supplied reporting does not establish that the benchmark setups, tasks, hardware, or cost calculations are directly comparable.
OpenAI has not released a public statement describing the reorganization, its detailed research plans, or how its robotics work will be evaluated. It is also unclear whether GPT-6 Astra’s reported performance will translate to real-world settings where delays, sensor variation, and physical constraints may affect control. The source material does not include the remainder of the NVIDIA study’s discussion of inference costs, so those details cannot be assessed here.
Evidence From Robots Outside Simulation
The next meaningful test will be whether the reported strengths hold across physical robots and varied tasks, rather than only in a benchmark or a company’s single-robot comparison. Further study details, independent testing of Praxis-1, and demonstrations from OpenAI could clarify how performance, response time, and operating costs compare.
OpenAI’s stated target for a first public humanoid-robot demonstration is 2027. The company has not provided a new timetable or said what role the reorganized teams will play in that effort. Until it releases more information, the scope of its robotics program and the practical relationship between general-purpose models and specialist policies remain open questions.
Key Questions
What changed in OpenAI’s robotics organization?
According to Fortune, OpenAI moved its robotics researchers into the main research team under Mark Chen and its applied robotics and hardware teams into the Applied organization under Greg Brockman. OpenAI confirmed the change to Fortune but has not issued its own public announcement.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter robot model released by Runway with weights and fine-tuning code under the Apache 2.0 license. Runway describes it as a world action model.
Did GPT-6 Astra outperform Praxis-1?
On RoboLab-120, Astra’s agentic setup scored 74.8%, compared with Praxis-1’s reported 46.2%. The scores come from different models and reported evaluation details; they do not establish which system is better across real-world robot tasks. Praxis-1 was reported to respond much faster in Runway’s separate test.
Has OpenAI abandoned robotics?
OpenAI’s spokesperson told Fortune that robotics remains part of its long-term vision and that the company remains committed to general-purpose robots. The team reorganization is confirmed, but OpenAI has not publicly explained its detailed plans following the change.
When might OpenAI demonstrate a humanoid robot?
CEO Sam Altman has targeted a public demonstration in 2027, according to the source material. That is a stated target; there is no confirmation here that a demonstration is scheduled or that a product will be available then.
Source: ThorstenMeyerAI.com
