🔍 Read the full analysis: Why OpenAI's Robotics Team Split And Praxis-1 Matter For Robot AI 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
OpenAI has disbanded its stand-alone robotics team, moving researchers into its main research organization and applied robotics and hardware teams into Applied. The company says general-purpose models can now be used directly for robot control, while Runway’s open-weight Praxis-1 results highlight a continuing trade-off: general models performed better in one simulation benchmark, but the specialist model acted much 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 staff into Applied, according to Fortune, which said OpenAI confirmed the change. The reorganization comes as Runway releases Praxis-1, an open-weight robot model, setting up a timely test of whether general-purpose AI can control robots as effectively as models built for that task.
Fortune reported November 13 that the robotics researchers now report to Chief Research Officer Mark Chen, while applied robotics and hardware teams have moved to the Applied organization under President Greg Brockman. OpenAI confirmed the reorganization to Fortune when asked. Its spokesperson described the move as a sign of continuing commitment to robotics, not a retreat. OpenAI had not published its own announcement as of November 14, according to the source report.
The company’s stated rationale is that its general-purpose frontier models, including GPT-6 Astra, have shown they can be applied to robot control directly. A study by NVIDIA researchers, published October 14 as a CoRL 2026 late-breaking result, tested Astra on RoboLab-120, a simulation benchmark of 120 tabletop tasks for a robot arm. In an agentic setup, it achieved a reported 74.8% success rate, compared with 51.0% for the best trained robot policy listed on the benchmark leaderboard.
That result came with a major speed limitation: in the study’s direct setup, each model call averaged 18.56 seconds, and the simulation paused while the model generated a response. Runway’s Praxis-1 scored 46.2% on RoboLab-120, the strongest result among the world action models and open-source robot models listed, and fifth overall. Runway says the 27-billion-parameter model takes about 0.55–0.6 seconds per call to produce 1.6 seconds of motion. These figures come from different systems and tests; they do not by themselves establish how either approach performs across real-world robots.
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 Is the Robot-Control Test
The competing results point to a practical distinction between solving tasks and acting in time. A model that succeeds more often in a paused simulation may still be too slow for a robot that must respond continuously to changing surroundings. Conversely, a faster specialist model may be less capable across a broad range of tasks. The published benchmark figures measure different aspects of that problem, not a definitive winner.
Runway also reported a comparison on one of its own robots: it 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 are Runway’s results, not an independent evaluation, and the source material does not specify enough about the test setup to generalize them to other robots or workloads.
The organizational choice matters because it signals how OpenAI plans to pursue robotics: by connecting robot work to its general-model research and applied teams rather than retaining a separate robotics group. For developers and researchers, Praxis-1 offers a contrasting approach, with weights and fine-tuning code released under Apache 2.0. The field may not need to choose one model for every job: broad models could handle planning or unusual tasks, while faster specialist policies handle immediate movement.
Products Worth Considering
LG UltraGear 32GX850A-B 32-inch 4K UHD OLED Gaming Monitor
DJI Mini 3 Drone with DJI RC and 3‑Axis Mechanical Gimbal
Samsung Galaxy Tab A11+ Kids Edition, 128 GB Android Tablet, Royal Blue
As an affiliate, we earn on qualifying purchases.
OpenAI’s Earlier Robotics Efforts
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 cited 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-maker Figure; Figure said in February 2025 that it would develop its own robot AI, ending that partnership.
OpenAI’s previous approach, as described by its spokesperson to Fortune, was to build on video models such as Sora, which model how the physical world changes. The company now says its frontier models can be applied to robotics directly. OpenAI has also said it is developing a humanoid robot, with CEO Sam Altman targeting a first public demonstration in 2027. The source report noted that the company’s CoRL booth remained on the floor plan but was empty when Fortune visited, consistent with a reorganization made after the booth had been registered.
“The two approaches are complementary.”
— Runway, describing Praxis-1
Real-World Performance Still Open
The evidence cited so far does not settle whether a general-purpose model can control physical robots reliably at useful speeds. NVIDIA’s Astra result is from a simulated benchmark, and the reported long call time is a substantial constraint in that setup. Praxis-1’s benchmark score and latency figures also do not establish performance across different robot bodies, environments, or task types.
The Runway comparison involved one of its robots and three tasks, and comes from the company itself. The available material does not provide independent replication or a full account of the test conditions. It is also unclear how OpenAI’s reorganized teams will divide responsibility for robot research, hardware, and deployment, or whether the company will publish a detailed account of the change.
Tests Must Move Beyond Simulation
The next useful evidence will be reproducible evaluations on physical robots that report success, response time, cost, and safety across comparable tasks. Further results could show whether GPT-6 Astra’s benchmark advantage persists when a robot must act without a paused simulation, and whether Praxis-1’s speed holds across other hardware and conditions.
OpenAI’s stated target for a public demonstration of its humanoid robot is 2027. Until then, the company has not set out a public timetable for the reorganized robotics work. Runway’s released weights and code give outside researchers a way to test and fine-tune Praxis-1, though the source material does not identify a shared evaluation that directly compares both approaches under the same real-world conditions.
Key Questions
What changed in OpenAI’s robotics organization?
According to Fortune, OpenAI moved its robotics researchers to the main research team under Mark Chen and its applied robotics and hardware teams to Applied under Greg Brockman. OpenAI confirmed the change to the outlet.
Why did OpenAI reorganize the team?
The company’s spokesperson told Fortune that general-purpose frontier models, including GPT-6 Astra, have shown they can be applied directly to robotics. OpenAI described the change as consistent with its continued commitment to robotics.
What is Praxis-1?
Praxis-1 is Runway’s 27-billion-parameter robot model, released with weights and fine-tuning code under the Apache 2.0 license. Runway reports a 46.2% score on RoboLab-120 and about 0.55–0.6 seconds per call for 1.6 seconds of motion.
Does the benchmark show that GPT-6 Astra is better than Praxis-1?
Not on its own. Astra’s reported 74.8% success rate came from an agentic simulation setup, while Praxis-1 scored 46.2% in the published benchmark results. Astra’s calls were much slower in the study’s direct setup, and the figures do not establish which model is better across physical robots or different tasks.
When might OpenAI show its humanoid robot?
CEO Sam Altman has targeted a public demonstration in 2027, according to the source material. OpenAI has not provided a more specific date in the reporting cited here.
Source: ThorstenMeyerAI.com
