Runway releases Praxis-1 world action model with Apache 2.0 weights
Runway's Praxis-1 is a 27B MoE world action model, and Runway says it finished robot tasks 15.6x faster than GPT-6 Astra.

Watch the short
Under a minute, read by an AI anchor. Made .
YouTube loads only when you press play.
Runway has released Praxis-1, a 27B mixture-of-experts diffusion-transformer world action model. Its weights are published under the Apache 2.0 license. Researchers posting about the release say it ranks first on RoboLab-120, a robotics benchmark.
The release comes with some infrastructure support. According to their posts, vLLM-Omni has day-zero support for the model. NVIDIA has published NVFP4 checkpoints along with a path to deploy on Jetson hardware.
The ranking claim and the speed claim come from different sources. Researchers report the RoboLab-120 ranking. The speed figures are attributed to Runway, which describes them as its own claims.
What changed
Runway has released Praxis-1, a 27B mixture-of-experts diffusion-transformer world action model, with weights published under Apache 2.0.
Type: New capability · Applies to: Praxis-1
Why it matters
Open Apache 2.0 weights give robotics teams a self-hostable alternative to closed world models, although the speed and cost gains are company-reported.
What you can do
Test the weights on your own robot arm or simulator setup before relying on the claimed speed and cost gains.
Original source
We have no link to the original source for this story yet.
Where we saw it
Latent Space
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works