🔍 Read the full analysis: Two Moves In Robotics AI: OpenAI's Team Split And Runway's Praxis-1 on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI has broken up its stand-alone robotics team, moving researchers into its main research organization and applied robotics and hardware staff into its Applied organization, according to Fortune. Runway has released Praxis-1, an open-weight robot model; available tests show a trade-off between broader simulated task performance and faster robot actions, not a settled winner.
OpenAI has broken up its stand-alone robotics team, moving researchers into its main research organization and applied robotics and hardware staff into its Applied organization, Fortune reported on November 13. The change came as Runway released Praxis-1, an open-weight model designed for robot control, bringing into focus whether robots should use general-purpose AI models or specialist systems.
Fortune said OpenAI confirmed the reorganization in response to its questions. The robotics researchers now report to Chief Research Officer Mark Chen, while the applied robotics and hardware teams moved to the organization under President Greg Brockman. An OpenAI spokesperson described the change as a sign of continued commitment to robotics, rather than a retreat. OpenAI had not published its own announcement about the move as of November 14.
According to Fortune, OpenAI said its reasoning changed as its general-purpose frontier models, including GPT-6 Astra, showed they could be applied directly to robotics. Earlier in the year, researchers had expected to build on video models such as Sora. The company’s spokesperson said robotics remains part of its long-term vision and that OpenAI remains committed to bringing general-purpose robots into the world.
Runway’s Praxis-1 has 27 billion parameters and is described as a “world action model.” It was released with weights and code for fine-tuning under the Apache 2.0 license, which permits commercial use. On the RoboLab-120 benchmark cited in the source material, Praxis-1 scored 46.2%, the highest among the world action models and open-source robot models listed there, 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 and Reach in Robot Control
The developments sharpen a practical trade-off for robotics developers: general models may handle a wider range of tasks, while specialist systems may respond quickly enough for direct control. In an NVIDIA researchers’ simulation study, an agent using GPT-6 Astra achieved a 74.8% success rate on RoboLab-120, above the best trained robot policy on that leaderboard. But in the study’s direct setup, each model call took an average of 18.56 seconds, and the simulation paused while it processed.
Praxis-1’s reported benchmark score was lower, but its calls took about 0.55 to 0.6 seconds while producing 1.6 seconds of motion. Runway also reported that, on one of its robots completing three tasks, 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, and should not be treated as a general result across robots or tasks.
For companies deciding how to build robot systems, the distinction matters because delays and inference costs can affect whether a model is practical to use repeatedly in the physical world. The results do not establish that one approach is better in every setting. Runway itself described the approaches as complementary, leaving open the possibility that a general model could plan while a faster specialist handles movement.
Popular right now
Roborock Qrevo S Pro Robot Vacuum and Mop, 18,500 Pa Suction
Vitamix VX1 Professional-Grade Blender, 64 oz, Black
Dyson V15 Detect Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
OpenAI’s Changing Robotics Structure
OpenAI’s team change is its second reported dissolution of a stand-alone robotics group. The company wound down its first robotics team in 2020. It rebuilt a team in 2024 and partnered with humanoid-robot maker Figure; Figure said the partnership ended in February 2025 as it pursued its own robot AI. Fortune also reported that OpenAI is developing a humanoid robot, with CEO Sam Altman targeting a public demonstration in 2027.
The timing of the latest reorganization was visible at the CoRL robotics conference in Austin. Although OpenAI was listed on the floor plan, Fortune found its booth empty. The account of the change comes from Fortune’s November 13 reporting; The Decoder’s November 14 report was based on that coverage. Fortune also cited pressure on OpenAI to control costs and focus on core products as business background, but the explanation the company gave for the reorganization was the capability of its general models.
The performance figures cited here come from different kinds of evidence. NVIDIA researchers’ RoboLab-120 result was a simulation study published October 14 as a CoRL 2026 late-breaking result. It tested GPT-6 Astra without robot-specific training. Praxis-1’s score and the comparative speed and cost figures are reported in the source material, with the latter attributed to Runway’s own test.
“Complementary.”
— Runway
What the Tests Do Not Establish
The available results do not settle whether general-purpose models or specialist robot models work better across real-world tasks. RoboLab-120 is a simulation benchmark, and its success rates do not by themselves show how either model performs across different robots, environments or safety requirements. The reported speed and cost comparison comes from Runway’s test on one of its robots and three tasks; the source material does not provide enough detail to establish how broadly those figures apply.
OpenAI has not published a detailed statement of its reorganization, and it is not clear how the new reporting lines will affect specific robotics projects, staffing or its humanoid-robot plans. It is also unclear whether OpenAI intends to release further real-world evaluations of GPT-6 Astra for robot control. The test results point to different strengths, but the evidence is not a direct, independent head-to-head comparison across identical conditions.
Further Tests and OpenAI Plans
Further comparisons will be needed to show how the two approaches perform on the same robots and tasks, including in real-world settings and under consistent measures of latency, cost and success. Praxis-1’s open weights and fine-tuning code allow developers to examine and adapt the model, though the source material does not specify any planned independent evaluations.
For OpenAI, the next indicators will be how the reorganized teams develop its robotics work and whether the company provides more detail about its robot plans. The source material cites a target for a public humanoid-robot demonstration in 2027; that is a stated target, not confirmation that a demonstration will occur. No further public timeline for the internal reorganization was given.
Key Questions
What changed in OpenAI’s robotics organization?
According to Fortune, OpenAI moved its robotics researchers into the main research organization under Mark Chen. Its applied robotics and hardware teams moved to the Applied organization under Greg Brockman.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter model for robot control. Runway released its weights and fine-tuning code under the Apache 2.0 license, which permits commercial use.
Which model scored higher on RoboLab-120?
The cited NVIDIA simulation study reported a 74.8% success rate for an agent using GPT-6 Astra. Praxis-1 scored 46.2% in the benchmark results reported in the source material. These results do not constitute a complete, independent head-to-head test under identical conditions.
Is Praxis-1 faster than GPT-6 Astra for robot control?
In Runway’s test on one of its robots and three tasks, Praxis-1 took about 0.55 to 0.6 seconds per call, while GPT-6 Astra took 15.6 times longer to finish the tasks. That comparison is attributed to Runway and may not apply to other robots or tasks.
Has OpenAI stopped working on robotics?
OpenAI’s spokesperson told Fortune that robotics remains part of the company’s long-term vision. The company reorganized its teams, but its plans and the effects of the change have not been fully detailed publicly.
Source: ThorstenMeyerAI.com
