Runway releases Praxis-1 robot model with open weights under Apache 2.0

Runway on 11 November made Praxis-1, its model for controlling robots, available to everyone with open weights. In its launch post, the company said the weights and the inference and fine-tuning code are released under the Apache 2.0 licence, including for commercial use. Runway first introduced Praxis-1 in September and says launch partners have been running it on a legged humanoid, a six-axis industrial arm and a dual-arm warehouse robot.

Runway calls Praxis-1 a world action model: it predicts what a scene will look like next and the motion needed to get there at the same time. The Runway research write-up says it is built on the company’s video foundation model, a mixture-of-experts diffusion transformer with about 27 billion parameters, of which roughly 2.8 billion are active per token, plus a separate action expert of about 314 million parameters. Each call returns a chunk of predicted video and 1.6 seconds of actions.

On performance, Runway says Praxis-1 tops NVIDIA’s RoboLab-120 leaderboard among world action models, reaching 46.2% success over 1,200 episodes against 42.9% for FLUX 3 Action and 28.0% for π0.5. It also reports 80.0% success on SimplerEnv after fine-tuning on Google’s RT-1 robot, which was not in its training data, compared with 52.4% for the original RT-1 policy. These are Runway’s figures.

The checkpoints are published in a Hugging Face collection that includes a base model plus versions for single-arm DROID and bimanual YAM rigs, LeRobot layouts and 4-bit NVFP4 variants. Runway says the model supports full fine-tuning and LoRA and runs on NVIDIA hardware from data centre GPUs down to Jetson Thor on the robot itself, with hosted inference also offered.

This report is not legal, financial, or health advice. See the disclaimer.