🔍 Read the full analysis: Is General AI Enough For Robots? OpenAI's Team Split And Runway's Praxis-1 on ThorstenMeyerAI.com
Get the little things that make your day 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 disbanded its stand-alone robotics team, moving researchers into its main research organization and applied robotics staff into its Applied group. Runway, meanwhile, released Praxis-1, an open-weight robot model. Published benchmark results suggest general models can succeed on more simulated tasks, while specialist models act faster; the companies’ approaches could also be combined.
OpenAI has disbanded its stand-alone robotics team and reassigned its researchers and applied robotics staff to other parts of the company, Fortune reported November 13, saying OpenAI confirmed the change when asked. The move came as Runway released Praxis-1, an open-weight model designed for robot control, putting renewed focus on whether general-purpose AI can operate robots effectively or dedicated systems are still needed.
According to Fortune, OpenAI’s robotics researchers now report to Chief Research Officer Mark Chen, while its applied robotics and hardware teams have moved to the Applied organization, led by President Greg Brockman. OpenAI’s spokesperson described the reorganization as a commitment to robotics, not a retreat. The company has not published its own announcement about the change, according to the source material.
The spokesperson told Fortune that earlier plans to build on video models such as Sora have shifted because general-purpose frontier models, including GPT-6 Astra, have shown they can be applied directly to robotics. A study by NVIDIA researchers tested Astra as a robot policy in simulation. Its agentic setup recorded a 74.8% success rate on RoboLab-120, a benchmark of 120 tabletop tasks. In the direct setup, however, model calls averaged 18.56 seconds, and the simulation paused while the model processed each call.
Runway’s Praxis-1 is a 27-billion-parameter “world action model” released with weights and code for fine-tuning under the Apache 2.0 license. Runway reports a 46.2% score on RoboLab-120, the highest among the world action models and open-source robot models on the cited leaderboard, and fifth overall. It takes about 0.55 to 0.6 seconds per call to produce 1.6 seconds of motion. In a separate test on one of its robots, Runway said GPT-6 Astra took 15.6 times longer and cost 51 times more per completed task than Praxis-1. Those figures come from Runway’s own test, not an independent 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
Speed and Skill Pull Apart
The results point to a practical trade-off for companies building robots: task success is not the only measure. A system that handles varied instructions or complex tasks in simulation may still be too slow or costly for a robot that must respond continuously in the physical world. Conversely, a faster specialist model may be less capable across a wide range of tasks.
That distinction matters because robot deployment depends on reliable action under time constraints, not just strong benchmark scores. If general models can provide planning or interpret instructions while smaller specialist systems handle immediate movements, developers may combine the two rather than choose one approach. The available results suggest that possibility, but do not establish how well it works outside the reported tests.
Products Worth Considering
Therabody Theragun Relief Massage Gun, 3-Speed, Navy
Philips Sonicare 5950 Rechargeable Electric Toothbrush with 2 Brush Heads, Pink
Withings Body Smart Wi-Fi and Bluetooth Body Composition Scale
As an affiliate, we earn on qualifying purchases.
OpenAI’s Robotics Reorganization
This is the second time OpenAI has closed a stand-alone robotics team. It wound down its first team in 2020; co-founder Wojciech Zaremba later cited limited robot data and a more efficient route toward AGI through other methods. 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 that partnership.
OpenAI has said it is developing a humanoid robot, and CEO Sam Altman has targeted a public demonstration in 2027. Fortune also reported that the company faces pressure to control costs and focus on core products, but that is background rather than the reason OpenAI gave for this reorganization. The stated explanation was that general-purpose models can now be applied directly to robotics.
The NVIDIA study cited in coverage was published October 14, 2026, as a CoRL 2026 late-breaking result. Its GPT-6 Astra tests were conducted in RoboLab-120 simulation, not in a general evaluation of robots operating in everyday settings. Praxis-1’s leaderboard result and Runway’s robot test are separate pieces of evidence, with different conditions.
“The two approaches are complementary.”
— Runway
Benchmarks Leave Practical Questions
The available figures do not establish which approach performs better across real-world robots, settings and tasks. NVIDIA’s 74.8% result came from a simulated benchmark, while Runway’s timing and cost comparison came from a test on one of its own robots. The source material does not provide enough detail to treat that single comparison as an independent, broadly representative result.
It is also unclear how OpenAI will organize future robotics research, whether it plans to publish further technical results, or how the reorganization affects development of its humanoid robot. The reported staff changes do not, by themselves, show that OpenAI has stopped building robot hardware. Nor do the studies settle whether a combined general-and-specialist system would deliver better performance in practice.
OpenAI’s 2027 Demonstration Target
The next public milestone to watch is the humanoid robot demonstration targeted for 2027 by Sam Altman. OpenAI has not specified in the reported material what the demonstration will include or how its newly organized teams will contribute.
Further evidence will depend on tests beyond simulation and company-run comparisons: results showing how models perform on physical robots, how quickly they respond, and what they cost to operate. Until such information is available, the reported benchmarks offer a snapshot of competing strengths, not a final answer on whether general AI is enough for robots.
Key Questions
What changed in OpenAI’s robotics organization?
Fortune reported that OpenAI moved robotics researchers into its main research organization under Mark Chen and applied robotics and hardware teams into its Applied organization under Greg Brockman. OpenAI confirmed the change to Fortune.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter model for robot action. Runway released its weights and fine-tuning code under the Apache 2.0 license, allowing commercial use under that license’s terms.
Does GPT-6 Astra outperform Praxis-1?
There is no single answer in the cited results. Astra scored 74.8% in an NVIDIA RoboLab-120 simulation setup, while Praxis-1 scored 46.2% on that benchmark. Praxis-1’s reported calls were much faster, and Runway’s separate test found Astra slower and more costly on one robot. The measures and test conditions differ.
Has OpenAI stopped developing robots?
The reported team reorganization does not establish that OpenAI has stopped robotics work. Its spokesperson said robotics remains part of the company’s long-term vision, and Altman has targeted a humanoid robot demonstration in 2027.
What remains unknown about the two approaches?
The reported results do not show how either model performs across a wide range of physical robots and real-world conditions. It also remains unclear whether OpenAI will publish further details about its reorganization or how the company plans to use general models in its robot development.
Source: ThorstenMeyerAI.com
