Runway announced on November 11, 2026 that Praxis-1, which it first introduced in September, is “now available to everyone with open weights under the Apache 2.0 license.”
What to know
- What it is: Runway calls Praxis-1 a world action model for robotics. It is a 27B-parameter mixture-of-experts diffusion transformer that predicts video and robot actions together. Runway says it learns from large-scale video pretraining and can be fine-tuned for new robots.
- Where to get it: weights and code on GitHub and Hugging Face (with native LeRobot support), serving through vLLM-Omni, hosted inference on Reactor, and a way to try it on Runway Dev.
- Hardware: Runway lists NVIDIA H100, B200, RTX 5090, RTX PRO 6000 and Jetson Thor.
- Licence: Apache 2.0, which Runway says includes commercial use. No price is given for hosted inference.
- Runway’s results: on NVIDIA’s RoboLab-120 leaderboard, Runway reports 46.2% success across 1,200 episodes, ahead of the other models it lists (42.9%, 36.8% and 28.0%). Fine-tuned for Google’s RT-1 robot, it reports 80.0% on SimplerEnv, against 52.4% for the original RT-1 policy. These are Runway’s own figures.
Why it matters
Runway is best known for creative video tools like Gen-4.5 and Aleph 2.0. Praxis-1 shows the same idea, a model that predicts what happens next in a video, being used to tell a robot what to do next. A permissive licence and support for a desktop GPU like the RTX 5090 make it one of the more accessible open robotics models. It is not a tool for making videos, so creators can keep using Runway’s video models.
More open models: the best open-weight AI models, what open-weight models are and the model tracker.
Source: Runway news. We summarise in our own words; check the source for the full details.
Contribute your creativity