Runway released on Wednesday, November 11, 2026, the open weights of Praxis-1, its artificial intelligence model for robotics, under the Apache 2.0 license — which includes commercial use. Introduced in September as the company's first "world action model", Praxis-1 turns video learning into commands for real robots.
The model is a 27-billion-parameter mixture-of-experts diffusion transformer that predicts video and actions together: it imagines the next frames of a scene and decides which movements the robot should make to get there. It builds on the same large-scale video pretraining behind Runway's general world models.
Performance on the RoboLab-120 leaderboard
According to Runway, Praxis-1 is the best-performing world action model on NVIDIA's RoboLab-120 leaderboard, with 46.2% success across 1,200 episodes — ahead of FLUX 3 Action (42.9%), Cosmos3 Nano (36.8%) and π0.5 (28.0%). The company also says it outperforms every listed open-weight vision-language-action (VLA) model. The figures come from Runway itself and have not yet been independently tested.
What is available
The release includes the weights, inference code and full and LoRA fine-tuning code, available on GitHub and Hugging Face with native LeRobot support, plus integration with vLLM-Omni and hosted inference on Reactor. There are three checkpoints: the base model, a version fine-tuned on the DROID dataset and one fine-tuned for the bimanual YAM robot.
The model runs on NVIDIA hardware, from data center GPUs such as the H100 and B200 to workstation cards such as the RTX 5090 and RTX PRO 6000, and even on Jetson Thor, inside the robot itself. At 54.4 GB in bfloat16, it fits on a single 80 GB GPU; with 4-bit NVFP4 quantization, the size drops to 17.5 GB.
Partners and adaptation
Three launch partners have already tested Praxis-1 on their own hardware: Noble Machines, with a legged humanoid; Standard Bots, with a six-axis industrial arm; and Ultra, with a dual-arm warehouse robot. Runway says a single checkpoint adapts to different robot bodies with light fine-tuning — and that, trained on Google's RT-1 robot, which never appeared in its training data, the model reached 80.0% success on SimplerEnv, versus 52.4% for the original RT-1 policy.
By opening the weights, Runway is keeping the commitment it made in its September announcement. The company argues that US leadership in physical AI depends on openness and interoperability, and points to its work as a founding member of the Cosmos Coalition, alongside NVIDIA.
Read also: Boston Dynamics puts scooter wheels on Atlas to test whole-body control
Image credit: official publicity image from Runway, published on the Praxis-1 introduction page.
Fontes
- Runway — Praxis-1 Is Now Available
- SXF Brief — AI News Nov 12, 2026
- MadRobot — Runway Praxis-1: free open-weight robot AI model released
- Daily AI Brief — Runway Releases Praxis-1 Robot Model With Open Weights
Transparency: this article was produced with the assistance of artificial intelligence, cross-checked against public posts on X and sources on the internet. We recommend consulting the original sources listed above to verify the information.
By Muse
