Runway unveils open-weight robotics model Praxis-1
Runway has announced Praxis-1, an open-weight world action model built on its video pretraining, moving the generative-video lab into robot control while the weights remain unreleased and partners test it.
Runway announced Praxis-1, which it describes as its first open-weight "world action model." The model turns the company's large-scale video pretraining into control policies for real robots. It is not yet publicly available. Runway says it will release the model with open weights "in the coming months."
The idea behind the model is that real-world robot data is scarce and expensive to collect, while video is effectively unlimited. Runway says a policy that has already learned physical plausibility and object behavior from video starts well ahead of one trained on action data alone. It compares this to language models learning from web text. The company reports that policy performance improves as it scales third-person video, and that web video matches teleoperated robot video as training data. It has not published head-to-head benchmarks against other policy models.
Runway also says that simulating robot policies inside its world model predicts real-world results with 0.95 correlation. It says this compares favorably with more expensive techniques based on 3D reconstruction. The figure is self-reported.
Praxis-1 builds on Runway's interactive world models, Solaris and GWM Worlds 2. Early partners Noble Machines, Standard Bots and Ultra are each running it on their own hardware. Runway says the pre-launch testing covers both efficacy and safety across different robot bodies and environments.
Runway explains its open-weight approach in terms of U.S. leadership in physical AI. It argues that American models need more interoperability and openness for physical use cases. The company has not given a release date, model size or licensing terms.