🔍 Read the full analysis: OpenAI And Runway Take Different Steps In The Race To Build AI For Robots on ThorstenMeyerAI.com
Get everyday essentials 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 moved its robotics researchers and applied robotics teams into other parts of the company, saying its general-purpose models can now be applied directly to robot control. Runway has taken a different approach, releasing the open-weight Praxis-1 for robotics. Early tests suggest general models can perform strongly in simulation, while specialist models respond faster; the comparison does not settle which approach will work best on real robots.
OpenAI has broken up its stand-alone robotics team and moved its researchers and applied robotics staff into other company groups, while Runway has released Praxis-1, a dedicated robot model with openly available weights. The contrasting moves, reported around the CoRL robotics conference in Austin, reflect different bets on whether general-purpose AI or models built specifically for robots will prove more useful for controlling machines.
Fortune reported on November 13 that OpenAI confirmed the reorganization after the outlet asked about it. Its robotics researchers now report to Chief Research Officer Mark Chen, while its applied robotics and hardware teams have moved into the Applied organization under President Greg Brockman. Fortune found the company’s registered CoRL booth empty. OpenAI has not published its own announcement about the change.
OpenAI’s spokesperson described the reorganization as a commitment to robotics, not a withdrawal. According to Fortune, the company said its general-purpose frontier models, including GPT-6 Astra, have shown they can be applied to robotics directly. The company had earlier expected to build on video models such as Sora. OpenAI has also said it is developing a humanoid robot, with CEO Sam Altman targeting a first public demonstration in 2027.
Runway’s Praxis-1 is a 27-billion-parameter “world action model” released under the Apache 2.0 license, with weights and code available for fine-tuning. In the reported RoboLab-120 results, it scored 46.2%, ranking first among the world action models and open-source robot models listed, and fifth overall. Runway’s figures put a call at about 0.55 to 0.6 seconds for 1.6 seconds of motion. Those measures come from specific benchmark and testing setups, not a broad demonstration of performance 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 Versus General Capability
The split matters because robot control places different demands on AI than answering a question or generating an image. A robot must select actions while responding to its surroundings. High task success in simulation may be useful, but slow responses can limit a model’s ability to control movement in real time.
A study by researchers at NVIDIA’s Seattle Robotics Lab illustrates that trade-off. In RoboLab-120, a simulation of 120 tabletop tasks for a robot arm, an agentic GPT-6 Astra setup achieved 74.8% success, compared with 51.0% for the best trained robot policy on the leaderboard cited in the source material. In the study’s direct setup, however, each model call took an average of 18.56 seconds, with the simulation paused while the model processed its response.
Runway says its own comparison on one of its robots found GPT-6 Astra took 15.6 times longer than Praxis-1 to finish three tasks and cost 51 times more per completed task. That is a company-reported result from a limited test, not an independent, general comparison. Taken together, the figures highlight the question facing developers: whether a general model’s broader capabilities can be paired with a faster control system, or whether specialist models will be needed for time-sensitive actions.
Products Worth Considering
Bose TV Speaker (2nd Gen) – Compact Soundbar with Bluetooth and HDMI‑ARC Connectivity, Clear Dialogue, Dolby Atmos, Deep Bass, Remote Contro
Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight
DJI Mini 5 Pro Fly More Combo with DJI RC 2
As an affiliate, we earn on qualifying purchases.
OpenAI’s Robotics Reorganization
OpenAI has previously changed course on robotics. It wound down an earlier stand-alone team in 2020; co-founder Wojciech Zaremba later cited limited robot data and a more efficient route to artificial general intelligence through other methods. The company rebuilt a robotics team in 2024 and partnered with humanoid maker Figure. Figure said the partnership ended in February 2025 as it moved to develop its own robot AI.
The latest change follows a shift in OpenAI’s stated technical rationale, according to the spokesperson’s account to Fortune: researchers had expected to build on video models, but the company now sees its general-purpose frontier models as directly applicable to robotics. Fortune also described cost control and focus on core products as part of the business backdrop. Those points are context, not the reason OpenAI gave for the team move.
The NVIDIA study was published on October 14, 2026, as a CoRL 2026 late-breaking result. It tested GPT-6 Astra as a robot policy that takes camera images and robot information and chooses movements without robot-specific training. The study reported 66.3% success for its direct setup, 74.8% for an agentic setup with persistent history and additional tools, and 66.5% for an orchestrated setup in which Astra reviewed movements proposed by a trained robot model.
“The two approaches are complementary.”
— Runway
Real-World Performance Still Unproven
The benchmark results are not a like-for-like verdict on which approach is better for all robots. RoboLab-120 is a simulation of tabletop tasks with one robot arm, and the GPT-6 Astra study’s reported success rates depend on the setup used. The source material does not provide an independent, broad real-world evaluation comparing GPT-6 Astra and Praxis-1 across the same robots, tasks and operating conditions.
Runway’s speed and cost comparison covers three tasks on one of its robots and is company-reported. It is not clear from the available material how the results would vary across different hardware, environments or longer task sequences. Nor has OpenAI publicly explained the team reorganization in a standalone statement; its rationale is reported through a spokesperson’s comments to Fortune. The companies’ longer-term plans for combining general and specialist models also remain unsettled.
CoRL Tests and 2027 Plans
The near-term test is whether the simulation findings translate into reliable control on physical robots. Further published results using shared tasks, hardware and measurement methods would make comparisons between general-purpose and specialist systems easier to assess. The source material does not identify a scheduled joint evaluation or a new public benchmark from either company.
OpenAI’s next stated public milestone is a planned first demonstration of its humanoid robot in 2027, according to the account of Sam Altman’s target in the source material. Whether OpenAI will release technical details about the reorganization or its robot-control work before then is not clear. Runway has made Praxis-1’s weights and fine-tuning code available, giving developers a route to test the model, but real-world performance across broader deployments remains to be established.
Key Questions
What did OpenAI change in its robotics organization?
According to Fortune, OpenAI moved its robotics researchers under Chief Research Officer Mark Chen and transferred its applied robotics and hardware teams to the Applied organization under President Greg Brockman. OpenAI confirmed the change when Fortune asked, but has not issued its own public announcement.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter robot model that Runway describes as a world action model. The company released it under the Apache 2.0 license, with weights and code available for fine-tuning and commercial use, according to the source material.
Does GPT-6 Astra outperform Praxis-1?
The reported results do not establish an overall winner. GPT-6 Astra scored 74.8% in an agentic RoboLab-120 simulation setup, while Praxis-1 scored 46.2% on that benchmark. The results come from different testing details, and the source material does not provide a broad, independent comparison on real robots.
Why might a dedicated robot model still be useful?
Robot actions can require fast responses. In the reported NVIDIA study, a direct GPT-6 Astra call averaged 18.56 seconds, while Runway reports Praxis-1 takes about 0.55 to 0.6 seconds per call for 1.6 seconds of motion. These figures come from different setups, but they show why response speed is part of the debate.
When could OpenAI demonstrate its humanoid robot?
OpenAI CEO Sam Altman has targeted a first public demonstration in 2027, according to the source material. The timing and scope of that demonstration have not been established in the supplied reporting.
Source: ThorstenMeyerAI.com
