Runway has announced that it is open-sourcing its core world action model, Praxis-1, under the Apache 2.0 license—releasing all model weights, inference code, and fine-tuning tools so that developers can freely use it in commercial projects. The model is a 27-billion-parameter mixture-of-experts diffusion Transformer. In the NVIDIA RoboLab-120 evaluation, it ranked first among comparable models with an overall success rate of 46.2%, and it can output both video predictions and action sequences simultaneously in a single inference pass. The model has been operationally validated on multiple real robotic hardware platforms of different form factors, demonstrating outstanding cross-embodiment transfer capability; after lightweight fine-tuning, it significantly outperformed native policies in previously unseen Google RT-1-related simulated environments. Its hardware compatibility is broad, covering full-chain deployment from cloud data center GPUs to edge chips on robot bodies, and it offers multiple inference invocation methods and fine-tuning pathways. The relevant resources have been hosted on an open-source platform, and Runway is working with industry partners to open up its frontier technology stack and promote the open development of the physical AI field.