ModelsHigh importance
Runway releases the weights of Praxis-1, a 27-billion-parameter model that uses video to control robots, under the Apache 2.0 license
Runway opened up Praxis-1 to everyone, after introducing it in September only for partners. It is a mixture-of-experts diffusion transformer with 27 billion parameters that predicts both the video and the robot’s actions from camera images, text instructions and the robot’s position. According to the company, it is the best “world action model” on NVIDIA’s RoboLab-120 ranking, with a 46.2% success rate across 1,200 episodes, and it beats every open-weight VLA model. Weights and inference code are released under Apache 2.0, which allows commercial use, on GitHub and Hugging Face (with LeRobot support), and also in vLLM-Omni. The Decoder notes that the model still confuses similar-looking objects and does not plan long sequences well on its own.
Published in edition no. 29 of Noticiário do Joãozinho ·
Translated by Joãozinho AI from the original Brazilian Portuguese edition. Facts, numbers and sources are the same as in the original; quotes, video titles and source names stay in their original language. Read the original in Portuguese
Original source
RunwayWeights on Hugging FaceCode on GitHubThe Decoder: the model’s limitations
https://runway.com/news/company-news/praxis-1-is-now-available