Runway announces Praxis-1, an open-weight world action model that turns its video pretraining into robot control

Runway has announced Praxis-1, its first open-weight world action model, aimed at controlling real robots. The company says it is built on the same large-scale video pretraining that underlies its general world models. Praxis-1 is being tested with early partners across a variety of robot embodiments. Runway says it will release the model publicly in the coming months. Runway's argument centers on a long-standing bottleneck in robotics. Real-world training data is scarce and expensive to collect at the volume a generalist policy needs. In deployments with frequent edge cases, such as autonomous driving or household robotics, the company says there is no practical way to gather enough. Video, by contrast, is abundant. People film far more of everyday life than any lab can capture through teleoperated demonstrations, and Runway expects generative video to expand that supply further. The company also reports that simulating robot policies inside its world model predicts real-world results with a 0.95 correlation. It says this compares favorably with more expensive techniques based on 3D reconstruction. If that holds up, developers could evaluate robot policies in simulation more cheaply before running physical trials. These figures are Runway's own, and the model has not yet been released for independent testing.