What happened
Runway, the company best known for AI video generation, released Praxis-1, a world action model for robots, with open weights on Wednesday, Nov. 11, as reported by MLQ.ai. The model is a 27-billion-parameter mixture-of-experts diffusion transformer that predicts video and robot actions together, built on the same video as Runway's general world models. Runway first announced Praxis-1 in September; the Nov. 11 release makes the weights and tooling publicly available.
According to MLQ.ai, Runway released Praxis-1 on Nov. 11 under an Apache 2.0 license, which permits commercial use. The release includes inference and code, with LoRA support so developers can adapt the model to their own robots, cameras, and tasks. Weights are available on GitHub and Hugging Face, with support for the LeRobot library and vLLM-Omni. MLQ.ai reports the model runs on Nvidia hardware ranging from H100 and B200 data center GPUs to RTX 5090 workstation cards and the Jetson Thor robot computer.
Runway reports performance figures that MLQ.ai attributes to the company, not to independent verification. On Nvidia's RoboLab-120 benchmark, Runway reports a 46.2% success rate across 1,200 episodes, ahead of FLUX 3 Action at 42.9%, Nvidia's Cosmos3 Nano at 36.8%, and Physical Intelligence's π0.5 at 28.0%. Runway calls it the top world action model on the leaderboard and says it beats every open-weight vision-language-action model; MLQ.ai notes that The Robot Report placed it fifth overall. Runway also reports that after on Google's RT-1 robot, which was not in its training data, Praxis-1 reached 80.0% on the SimplerEnv test versus 52.4% for the original RT-1 policy.
MLQ.ai reports three launch partners: Noble Machines, which builds a legged humanoid for heavy industrial handling; Standard Bots, which makes six-axis arms for machine tending and palletizing; and Ultra, which deploys dual-arm stations that pack warehouse orders. In a Noble Machines warehouse task, Runway says the model learned from half an hour of examples to place wooden planks in a container 15 times out of 15, and cleared an unfamiliar mixed pile in 6 of 10 attempts. These partner results are Runway's own claims and have not been independently confirmed.
Why it matters
The release puts a commercially usable, open-weights robot control model in developers' hands at no license cost, with support that lowers the barrier for adapting the model to specific robots and tasks. It also marks a notable expansion of Runway beyond video generation into physical AI, entering a field where Nvidia, Physical Intelligence, and others are already competing.
The Apache 2.0 license and included tooling mean robotics developers can commercially use and adapt the model without negotiating a license, which is a materially different proposition from closed robot foundation models. The LoRA support and compatibility with the LeRobot library and consumer-adjacent hardware like the RTX 5090 and Jetson Thor widen the pool of teams that can realistically experiment with it.
The release also signals Runway's move from video generation into physical AI, a field MLQ.ai describes as crowded: Nvidia has opened four Cosmos 3 models for robot simulation, and Nvidia and Unity backed Emulate's $700 million round for world models. Runway's head of robotics, Andy Chen, framed the decision in a statement quoted by MLQ.ai: "We view open world models as a compounding advantage."
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Which component of an AI application is the machine-learning model itself?
What to watch next
Whether independent teams replicate Runway's reported benchmark results, whether the three launch partners deploy the model in production settings, and how the model performs on hardware and robots outside Runway's tested configurations.
Independent benchmark replications. All headline numbers, including the RoboLab-120 and SimplerEnv results and the partner warehouse demonstrations, are Runway-reported and have not been independently verified.
Real-world deployments by Noble Machines, Standard Bots, and Ultra, which would test whether the reported warehouse performance holds outside demonstration conditions.
Community fine-tunes on robots and hardware not covered in Runway's release, and how quickly competing open-weight and closed robot models respond on the same benchmarks.