New to this? Read it in simple words
- Runway, a video AI company, released its robot model Praxis-1 for anyone to download.
- The model learns from video how things move, then tells a robot how to move.
- Runway says it beat FLUX 3 Action on a robot test, 46.2% against 42.9%.
- Runway ran the tests itself, and nobody outside has checked them yet.
- Open weights
- The trained model is published, so anyone can download, run and change it.
- World action model
- An AI that predicts how a scene will change and chooses the movements that get there.
- Simulation
- A computer copy of the real world, used to test robots without real machines.
Runway is best known for AI that makes video. With Praxis-1, it uses the same video research to control robots, and now anyone can download the model.
What Praxis-1 does
Praxis-1 takes in camera images, a text instruction and the robot’s position. It predicts how the scene will change and works out the robot movements to get there. As new camera frames arrive, it keeps adjusting.
Runway calls it a world action model. It has 27 billion parameters and is built on Runway’s own video model, which learned how objects behave from ordinary footage.
“A model trained to predict video really understands a lot about the physical world,” said Andy Chen, Runway’s head of robotics, as The Robot Report reports.
RoboLab-120 success rates, all as Runway reports them. Simulation only.
Who can use it
The weights are on Hugging Face and the code is on GitHub. Runway says both are under the Apache 2.0 licence, which allows commercial use. The Decoder says users must sign up and accept access terms on Hugging Face.
The full model needs a graphics chip with 80 GB of memory, such as one NVIDIA H100. A compressed version should fit on an RTX 5090 with 32 GB, Runway estimates, but its peak memory use has not been measured yet.
The base model needs extra training before it works on a new robot. Runway also offers versions already adapted to some robots.
How it scored
On NVIDIA’s RoboLab-120 test, which runs 120 tasks in simulation, Runway says Praxis-1 succeeded in 46.2% of 1,200 tries. It names 42.9% for FLUX 3 Action, 36.8% for Cosmos3 Nano and 28.0% for π0.5.
That makes it the top world action model on the list. Overall it ranks fifth, The Robot Report and The Decoder note. Runway ran all the tests itself, and nobody outside has checked them yet.
Runway also trained the model on Google’s RT-1 robot, which was not in its earlier training data. In simulated tests it then reached 80.0%, against 52.4% for Google’s original RT-1 model, Runway says.
Limits and partners
Runway says the model sometimes confuses objects that look alike and struggles with vague instructions. It cannot yet plan long tasks on its own, so each step must be spelled out, The Decoder reports from Runway’s research report.
Three robot makers have tested it since September: Noble Machines with a humanoid, Standard Bots with an industrial arm, and Ultra with a two-armed warehouse robot. Chen says that on one warehouse task, the model learned from half an hour of examples and got it right 15 times out of 15.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2Research · November 12, 2026Runway’s Praxis-1 world action model for robotics is now available The Robot Report
- 3Research · November 14, 2026Runway’s open-source Praxis-1 turns video AI into a robot controller The Decoder
- 4
- 5Research · November 12, 2026Runway has unveiled Praxis-1, an AI that understands the world from videos and controls robots GIGAZINE
All test scores are Runway’s own, and no outside group has repeated them. The ranking on the overall list comes from The Robot Report and The Decoder.
Runway opens its Praxis-1 robot model to everyone. Silicon AI News, November 14, 2026. https://siliconainews.com/stories/runway-praxis-1-open-weights
Runway made its 27-billion-parameter robot model free to download for commercial use, though its test wins are self-reported. How we rate the mood
The mood shows how the news frames the event for the tech and AI world: as an opportunity (6 and above) or as a loss (below 5). It is not our verdict on who is right.