Runway made Praxis-1 generally available on 11 November, releasing its robot model with open weights under the Apache 2.0 licence โ including commercial use. Praxis-1 is a 27-billion-parameter mixture-of-experts diffusion transformer that Runway calls a "world action model": it predicts video and robot actions together, learning how the physical world moves by watching video rather than only from robot demonstration data.
On NVIDIA's RoboLab-120 benchmark, Praxis-1 succeeded in 555 of 1,200 episodes โ a 46.2% success rate โ ahead of FLUX 3 Action at 42.9%. Those are Runway's own numbers, not independently verified, so treat the lead as the company's claim until third parties reproduce it.
The release is genuinely open: weights, inference code, and both full and LoRA fine-tuning code are on GitHub and Hugging Face, with native LeRobot support and vLLM-Omni serving. It runs on H100 and B200 data-centre GPUs, RTX 5090 and PRO 6000 workstation cards, and NVIDIA's Jetson Thor edge platform โ a wide enough range to cover both labs and production robots.
Launch partners Noble Machines, Standard Bots and Ultra have already run Praxis-1 on real machines. The model was first announced in September; the public release on 11 November turns it from a research preview into something robot builders can download, fine-tune and ship today.