Product Launch read

Runway releases Praxis-1 robotics model with open weights

Runway released Praxis-1, a 27B-parameter world action model for robots, under the Apache 2.0 license. Runway says it tops NVIDIA's RoboLab-120 leaderboard for world action models with a 46.2% success rate.

AE
Aidose Editorial
Nov 14, 2026

Runway has released Praxis-1, its first world action model for robotics, as open weights under the Apache 2.0 license. The company announced the release on November 11, and the license allows commercial use. Runway first introduced the model in September.

Praxis-1 is a 27-billion-parameter mixture-of-experts diffusion transformer that predicts video and robot actions together. It is built on the same large-scale video pretraining Runway uses for its general world models. The release includes the weights plus inference and fine-tuning code, with support for both full fine-tuning and LoRA.

On NVIDIA's RoboLab-120 leaderboard, which runs 1,200 episodes, Runway reports Praxis-1 at 46.2% success, ahead of FLUX 3 Action at 42.9%, Cosmos3 Nano at 36.8% and π0.5 at 28.0%. On SimplerEnv, Runway says a version fine-tuned for Google's RT-1 robot reached 80.0% success, compared with 52.4% for the original RT-1 policy.

The model runs on NVIDIA GPUs from data center cards such as the H100 and B200, to workstation cards including the RTX 5090 and RTX PRO 6000, to the Jetson Thor robot computer. Code is on GitHub, weights are on Hugging Face with native LeRobot support, and the model also works in vLLM-Omni. Runway offers hosted inference through Reactor.

Because the model ships with a permissive license and fine-tuning code, robotics teams can adapt it to their own hardware without licensing negotiations. Runway, best known for video generation, now has an open model in robotics, where open-weight action models have been scarce.

AE
Aidose Editorial
Nov 14, 2026 · read
Back to News