Runway releases Praxis-1, its 27.2-billion-parameter robot model, as open weights under Apache 2.0, and reports a higher RoboLab-120 score than the four models it lists
By the AI Switchboard · researched and written with AI from the sources listed on this page, each point checked against its source before publishing · how we work · report a mistake
The AI video company introduced the model in September; weights and code are now free for commercial use.
Runway, best known for AI video tools, made Praxis-1 freely available on 11 November. It is what Runway calls a world action model: it predicts video frames and robot actions together. Runway's research page gives 27.2 billion parameters, with a mixture-of-experts video backbone of about 2.8 billion active parameters per token and an action module of about 314 million. The weights and the inference and fine-tuning code are under the Apache 2.0 licence, commercial use included, on GitHub, Hugging Face and vLLM-Omni.
Runway's own scores, set against the comparisons it lists: 46.2% success on the RoboLab-120 benchmark over 1,200 episodes, against 42.9% for FLUX 3 Action, 36.8% for Cosmos 3 Nano, 28.0% for π0.5 and 7.3% for Nvidia's GR00T N1.6. Fine-tuned on Google's RT-1 robot, which it was not trained on, it reaches 80.0% on SimplerEnv, against 52.4% for the original RT-1 policy. These are Runway's measurements. The Robot Report says Praxis-1 ranks fifth overall on the RoboLab-120 leaderboard, so the models it lists are not the whole field.
Two steps: Runway introduced Praxis-1 in September, and on 11 November opened it to everyone. Its launch partners include Noble Machines, Standard Bots and Ultra; The Robot Report describes them as makers of a legged humanoid, six-axis arms and dual-arm warehouse stations. The open release means a small robot maker can now fine-tune a large robot model without licensing one.
“We believe the frontier of robotics will be discovered through real-world applications, not in a lab,” Andy Chen, Runway's head of robotics, told The Robot Report.
- Confirmed Praxis-1's weights and code are under Apache 2.0, commercial use included, on GitHub, Hugging Face and vLLM-Omni. Runway
- Claimed Runway reports 46.2% on RoboLab-120 against 42.9% for FLUX 3 Action and 36.8% for Cosmos 3 Nano, and 80.0% on SimplerEnv against RT-1's 52.4%. Runway
- Reported Praxis-1 ranks fifth overall on the RoboLab-120 leaderboard, The Robot Report says. The Robot Report
Robots & physical AIModels & releases
Today in the November 12, 2026 edition · front page