🔍 Read the full analysis: Runway Opens Praxis-1 As OpenAI Rethinks Its Robotics Team on ThorstenMeyerAI.com
Get hardware and tech 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
Runway released Praxis-1, an open-weight model designed for robot control, two days before OpenAI confirmed it had reorganized its stand-alone robotics team. OpenAI says its general-purpose models can now be applied directly to robotics; published results point to a trade-off between task success and response speed.
Runway has released Praxis-1, an open-weight model designed to control robots, as OpenAI reorganizes its stand-alone robotics team and places its researchers in broader company groups. The contrasting moves sharpen a question for robotics developers: whether general-purpose AI models can handle robot control, or whether robots still need models built specifically for physical actions.
Fortune reported on November 13 that OpenAI had broken up its dedicated robotics team and moved staff into other parts of the company. OpenAI confirmed the reorganization when Fortune asked. Its research team now reports to Chief Research Officer Mark Chen, while applied robotics and hardware teams moved to the Applied organization under President Greg Brockman.
OpenAI’s spokesperson told Fortune the change reflects the company’s view that its general-purpose frontier models, including GPT-6 Astra, can be applied to robotics directly. The spokesperson described robotics as part of OpenAI’s long-term vision and said the company remains committed to bringing general-purpose robots into the world. OpenAI has not published its own announcement about the reorganization.
Runway released Praxis-1 two days before the Fortune report. The model has 27 billion parameters and was published with weights and code for fine-tuning under the Apache 2.0 license, which permits commercial use subject to the license’s terms. Runway calls it a “world action model.” On the RoboLab-120 benchmark, it recorded a 46.2% success rate, according to the figures in the source material.
A separate simulation study by NVIDIA researchers tested GPT-6 Astra on the same benchmark. Its agentic setup reached 74.8% task success, while a direct setup reached 66.3%. The study’s authors said the agentic result exceeded the benchmark leaderboard’s best policy by 23.8 percentage points. Those results do not settle how the systems perform on physical robots: the benchmark consists of simulated tabletop tasks.
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 Success Pull Apart
The releases highlight a practical trade-off in robot AI: completing more tasks is not the same as acting quickly enough to control a robot in real time. In NVIDIA’s direct GPT-6 Astra setup, a model call took an average of 18.56 seconds, and the simulation paused while the model generated its response. Praxis-1’s reported calls took about 0.55 to 0.6 seconds for 1.6 seconds of motion, according to the source material.
That difference matters because a robot must respond to changing surroundings, not merely produce a successful plan eventually. Runway also reported that, in its test 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. These are Runway’s own results, not an independent comparison, and the source does not provide enough detail to establish how broadly they apply.
The developments may also affect how robotics companies allocate research effort. OpenAI’s reorganization puts robotics researchers closer to teams building its general models, while Runway is making a robot-focused model available for others to download and adapt. The evidence so far supports neither a conclusion that one approach has won nor that specialist models are no longer needed.
Products Worth Considering
LG UltraGear 24-Inch Full HD IPS Gaming Monitor, 120Hz (144Hz Overclocked), 1ms MBR, G-SYNC Compatible, FreeSync, HDR10
Ledger Stax Crypto Wallet with Touchscreen
LG UltraGear evo 27-inch 5K Mini-LED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
OpenAI’s Second Robotics Reorganization
The current reorganization is OpenAI’s second move away from a stand-alone robotics team. The company wound down an earlier robotics effort in 2020. It rebuilt a team in 2024 and partnered with humanoid robot maker Figure; Figure said in February 2025 that it would develop its own robot AI, ending that partnership.
OpenAI’s explanation for the latest change is a shift in its technical assessment. According to its spokesperson, researchers had expected earlier in the year to build on video models such as Sora, but later concluded that general-purpose frontier models could be applied to robot control directly. Fortune also cited pressure on OpenAI to rein in costs and focus on core products as business background; the company’s stated reason for the change was its view of the models’ capabilities.
The NVIDIA research cited in the reports was published on October 14, 2026, as a CoRL 2026 late-breaking result. It tested GPT-6 Astra as a “native robot policy,” giving it camera images, robot information and a control interface without robot-specific training. The test used RoboLab-120, a simulation of 120 tabletop tasks involving one robot arm—not a trial of general-purpose humanoid robots in everyday settings.
“The two approaches are “complementary.””
— Runway
Simulation Results Leave Gaps
Real-world performance remains an open question. The NVIDIA result comes from simulation, and the supplied information does not establish whether the 74.8% success rate transfers to physical robots or to tasks beyond the benchmark. The source also does not provide comparable independent testing of GPT-6 Astra and Praxis-1 across a common set of real-world conditions.
The speed and cost comparison comes from Runway’s own test on one of its robots involving three tasks. The full test protocol, task definitions and wider sample are not specified in the material provided, so the reported 15.6-times time difference and 51-times cost difference should not be treated as general results. OpenAI has not publicly detailed the reorganization, its staffing implications or how robotics work will be measured going forward.
It is also unclear whether OpenAI’s stated interest in general-purpose robots includes a particular development schedule beyond CEO Sam Altman’s previously reported target of a public demonstration in 2027. The source material does not confirm whether that target has changed.
Tests Will Shape the Next Step
The next useful evidence will come from comparable tests on physical robots, including response times, costs and performance across a broader range of tasks. Researchers and developers will also need to show whether general-purpose models can operate reliably without pausing action while they reason, and whether combining them with faster specialist policies improves results.
OpenAI’s next public signals may clarify how its newly distributed robotics teams will work and whether the company will share further details about its robot program. The source material reports a target for a public demonstration in 2027, but provides no new schedule or confirmation of a demonstration. Runway’s open release gives outside developers access to Praxis-1’s weights and fine-tuning code; their results may offer a broader test of the specialist-model approach.
Key Questions
What changed at OpenAI?
Fortune reported that OpenAI moved its robotics researchers into the main research organization under Mark Chen and shifted its applied robotics and hardware teams to the Applied organization under Greg Brockman. OpenAI confirmed the reorganization to Fortune.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter model designed for robot control. Runway released its weights and fine-tuning code under the Apache 2.0 license, which permits commercial use subject to the license’s terms.
Did GPT-6 Astra outperform Praxis-1?
In the cited RoboLab-120 simulation results, GPT-6 Astra’s agentic setup scored 74.8%, while Praxis-1 scored 46.2%. The figures come from different sources and do not establish that Astra is better for real-world robot control; the reported speed tests point to a substantial latency difference.
Are the models proven to work on real-world robots?
No broad real-world conclusion is established by the material provided. NVIDIA’s reported benchmark tested simulated tabletop tasks, while Runway’s time and cost comparison was its own test on one of its robots across three tasks.
What happens next for OpenAI’s robotics work?
OpenAI has said robotics remains part of its long-term vision, and CEO Sam Altman has previously targeted a public robot demonstration in 2027. The company has not provided a new schedule or detailed a public plan for the reorganized teams.
Source: ThorstenMeyerAI.com
