Runway's open-source Praxis-1 turns video AI into a robot controller
Reported by The Robot Report and The Decoder2 sources ↓
Runway released Praxis-1’s weights and source code under the Apache 2.0 license. In company-run tests on Nvidia’s RoboLab-120, it led the World Action Model category; results have not been independently verified.
Key points
- Praxis-1 scored 46.2% on RoboLab-120, ahead of FLUX 3 Action at 42.9% in the same category
- On a Franka robot arm, Praxis-1 completed 88 of 100 attempts across 10 tasks; FLUX 3 Action completed 80
- Developers can access the model through Hugging Face and its code through GitHub
- The base model needs adaptation for new robot systems; its standard version requires an 80-GB GPU
Sources
Details
Praxis-1 takes camera images, text instructions and a robot’s position as inputs, then predicts changes in the scene and calculates robot movements. It is based on a video model fine-tuned with human-action videos and robot data, and adjusts its control as new camera frames arrive.
Runway reports that Praxis-1 scored 46.2% across 120 simulated tasks on Nvidia’s RoboLab-120, compared with 42.9% for FLUX 3 Action. It ranked fifth overall, behind two AI agents and two robot models with different architectures. Runway conducted the tests, and the results have not been independently verified.
Runway also reports that Praxis-1 succeeded in 88 of 100 attempts on a real Franka robot arm, while FLUX 3 Action succeeded in 80. For a robot type absent from its earlier training, Praxis-1 reached an 80% average in simulated tests after fine-tuning on about 95 hours of robot data; Google’s RT-1 scored 52.4% on that benchmark.
The weights are available through Hugging Face after users sign up and agree to access terms; the code is on GitHub. Runway says pre-adapted versions for some robots are available, while the base model needs additional training for new systems. The company’s report says Praxis-1 can confuse similar-looking objects, has difficulty with vague instructions and cannot reliably plan complex sequences on its own.
Written from coverage as of
Images
- The Decoder
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh