🔍 Read the full analysis: OpenAI's Robotics Team Reorganizes As Runway Opens Praxis-1 on ThorstenMeyerAI.com
Get smart everyday buys 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
Fortune reported that OpenAI has ended its stand-alone robotics team, moving researchers into its main research organization and applied robotics and hardware staff into its Applied organization. The change comes as Runway releases Praxis-1, an open-weight robot model; published results point to a trade-off between general models’ simulation performance and specialist models’ speed.
OpenAI has broken up its stand-alone robotics team, moving researchers into its main research organization and applied robotics and hardware teams into its Applied organization, Fortune reported November 13. The change coincides with Runway’s release of Praxis-1, an open-weight robot model, sharpening a debate over whether robots need specialized AI or can be controlled by general-purpose models.
Fortune said OpenAI confirmed the reorganization when asked. Its robotics researchers now report to Chief Research Officer Mark Chen, while applied robotics and hardware teams have moved into the Applied organization, which reports to President Greg Brockman. OpenAI had registered a booth at the CoRL robotics conference in Austin, but Fortune found it empty. OpenAI had not posted a statement about the change on its own news feed as of November 14.
According to Fortune, an OpenAI spokesperson said the shift reflects the company’s view that general-purpose frontier models, including GPT-6 Astra, can be applied directly to robotics. The spokesperson described robotics as a core part of OpenAI’s long-term vision and said the company remains committed to bringing general-purpose robots into the world. OpenAI’s stated explanation is about how it will pursue robotics, not that the field no longer matters.
Runway released Praxis-1 two days before the report. The company describes it as a 27-billion-parameter “world action model” and released its weights and fine-tuning code under the Apache 2.0 license, which permits commercial use. On RoboLab-120, Praxis-1 scored 46.2%, according to the reported results. That was the highest score among the world action models and open-source robot models on the leaderboard, 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 May Favor Specialist Models
The reorganization and the release land amid a practical question for robotics developers: can a broad AI model plan and control physical actions well enough, or does a robot need a model built specifically for fast movement? Available results suggest there is no single winner across measures. General models have posted stronger success rates in some simulations, while specialist robot models have a substantial speed advantage in the reported tests.
A study by researchers at NVIDIA’s Seattle Robotics Lab tested GPT-6 Astra on RoboLab-120, a simulation of 120 tabletop tasks for one robot arm. In an agentic setup, Astra achieved a 74.8% success rate, compared with 51.0% for the best trained robot policy on the cited leaderboard. But in the study’s direct setup, a model call took an average of 18.56 seconds, and the simulation paused while Astra processed the request. Those conditions limit what the success figure says about performance in a physical setting where a robot must respond continuously.
Runway reported a different comparison on one of its robots: GPT-6 Astra took 15.6 times longer than Praxis-1 to complete three tasks and cost 51 times more per completed task. That is a company-reported test involving one robot and three tasks, not a broad independent evaluation. Together, the results highlight trade-offs that could shape deployment costs, response times and system design, rather than settling which model approach is best.
Products Worth Considering
Logitech MX Keys S Combo Wireless Keyboard and Mouse, Graphite
Dell UltraSharp U4025QW 40 5K2K WUHD Curved LED Monitor (21:9)
Canon SELPHY CP1500 Wireless Dye-Sublimation Photo Printer Bundle with 108 Sheets of Paper and 3 Ink Cartridges
As an affiliate, we earn on qualifying purchases.
OpenAI’s Robotics Team History
OpenAI previously wound down a stand-alone robotics team in 2020. Cofounder Wojciech Zaremba later cited limited robot data and the prospect of pursuing artificial general intelligence more efficiently through other methods. The company rebuilt a robotics team in 2024 and partnered with humanoid robot maker Figure. Figure said the partnership ended in February 2025 as it moved to develop its own robot AI.
The latest reorganization follows an earlier OpenAI research assumption, described by its spokesperson to Fortune: that robotics work would build on video models such as Sora, which generate video by modeling how the physical world behaves. The company now says its general-purpose frontier models can be applied directly, a change that helps explain why its researchers have been placed within broader research and applied organizations.
OpenAI has said it is developing a humanoid robot, and CEO Sam Altman has targeted a public demonstration in 2027. Fortune also cited pressure on OpenAI to control costs and focus on core products as business background. The outlet did not report that cost pressure was the company’s stated reason for the reorganization; OpenAI’s explanation was the potential of general models for robotics.
“The two approaches are “complementary.””
— Runway
Real-World Performance Remains Open
The cited GPT-6 Astra result comes from simulation, not a demonstration of the model reliably controlling robots in varied physical environments. The study’s average 18.56-second call time also reflects its direct setup, where the simulation paused during processing; it should not be treated as a universal response time for every configuration.
Runway’s speed and cost comparison covers three tasks on one of its robots, and the source material does not provide enough detail to establish how those results would hold across different robots, tasks or independent tests. The available figures also use different setups, so they do not amount to a head-to-head evaluation under one shared protocol. OpenAI has not published its own detailed account of the reorganization, and the precise division of work across its new organizations remains unclear.
It is also not yet clear whether OpenAI will build a dedicated robot model inside its broader teams, how its stated humanoid robot plans fit with the reorganization, or whether general and specialist models will be combined in deployed systems. The benchmark results offer evidence about particular tests, not a settled answer about which approach will work best in commercial robots.
Watch for Demos and New Benchmarks
The next useful evidence will come from tests outside simulation, with comparable measures of task success, response time and cost across multiple robots and tasks. Further details from OpenAI about the new team structure could clarify how research on models will connect to hardware and applied robotics work.
OpenAI’s stated target for a public demonstration of its humanoid robot is 2027. Until then, its plans and the reorganization’s practical effects remain developing matters. Runway’s open weights and fine-tuning code also allow outside developers to test Praxis-1 commercially, but broader results are not established by the release or the small comparison described so far.
Key Questions
What changed in OpenAI’s robotics organization?
According to Fortune, OpenAI ended its stand-alone robotics team. Robotics researchers moved into the main research organization under Mark Chen, while applied robotics and hardware teams moved into the Applied organization under Greg Brockman.
Why did OpenAI reorganize its robotics team?
An OpenAI spokesperson told Fortune that general-purpose frontier models, including GPT-6 Astra, have shown they can be applied directly to robotics. The company framed the reorganization as a change in approach, not a retreat from robotics.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter robot model that Runway calls a “world action model.” Runway released its weights and fine-tuning code under the Apache 2.0 license, which permits commercial use.
Do the results show that general models are better at robotics?
Not conclusively. GPT-6 Astra achieved a higher success rate than the cited trained policy in one simulation setup, while the reported tests also show that specialist models can act much faster. The studies use different conditions and do not establish a universal winner.
What remains unknown about OpenAI’s robotics plans?
OpenAI has not published a detailed statement describing the reorganization, and it is unclear how its humanoid robot plans will fit with the new team structure. Its CEO has targeted a public demonstration in 2027, but the cited results do not establish how the model approaches will perform across physical robots and tasks.
Source: ThorstenMeyerAI.com
