Runway Releases Praxis-1 Robot Model With Open Weights

Runway released the weights and code for its Praxis-1 world action model under the Apache 2.0 license, the company said in an announcement. Runway says the model recorded a 46.2% success rate across 1,200 episodes on NVIDIA's RoboLab-120 benchmark, compared with 42.9% for FLUX 3 Action and 36.8% for Cosmos3 Nano.

Praxis-1 is a 27 billion parameter mixture of experts diffusion transformer that predicts video and actions together. It is available through GitHub, Hugging Face with native LeRobot support, vLLM-Omni and hosted inference service Reactor. The release includes inference and fine tuning code, with support for full fine tuning and LoRA.

The model supports NVIDIA hardware including H100 and B200 data center GPUs, RTX 5090 and RTX PRO 6000 workstation GPUs, and Jetson Thor. Runway also tested Praxis-1 on Google's RT-1 robot, where a fine tuned version reached an 80% success rate in SimplerEnv, compared with 52.4% for the original RT-1 policy.

Runway introduced Praxis-1 in September as its first world action model. Launch partners Noble Machines, Standard Bots and Ultra have been running it on a legged humanoid, a six axis industrial arm and a dual arm warehouse robot.