Video AI Company Runway Gives Away a Robot Brain: Praxis-1 Tops Nvidia's Robotics Leaderboard and Is Free for Commercial Use

Most people know Runway for AI-generated video. Now it has released an AI model that controls real robots, free for anyone to use, including commercially.

What happened

On Nov. 11, Runway announced on its website and on X that its robotics model Praxis-1 is now available to everyone with open weights under the Apache 2.0 license. That means anyone can download the model, the inference code and the fine-tuning code, run and modify them on their own hardware, and use them commercially. Runway first introduced Praxis-1 in September and had limited it to launch partners: a legged humanoid from Noble Machines, a six-axis industrial arm from Standard Bots and a dual-arm warehouse robot from Ultra.

What it is

Praxis-1 is what Runway calls a "world action model": a 27-billion-parameter mixture-of-experts diffusion transformer that predicts both what the scene will look like next and how the robot should move. A rough analogy: traditional robot software follows a manual, while this model first watches huge amounts of video to get a feel for how objects get picked up, how hands move and what a half-finished task looks like, and then learns specific robot motions.

Runway's numbers, all self-reported: on Nvidia's RoboLab-120 leaderboard it ranks first with a 46.2% success rate across 1,200 episodes, ahead of FLUX 3 Action at 42.9%, Cosmos3 Nano at 36.8% and π0.5 at 28.0%. Fine-tuned on Google's RT-1 robot, which never appeared in its training data, it reached 80.0% success on SimplerEnv versus 52.4% for the original RT-1 policy. Runway also says that with under 30 minutes of teleoperated demos it succeeded in all 15 simulated trials on Noble Machines' robot, and that on the same station and tasks it finished 15.6 times faster than GPT-6 Astra driving the arm, at 51 times lower cost per completed task.

It runs on Nvidia hardware from data center GPUs down to the Jetson Thor chip on the robot itself, and supports LoRA, a lightweight way to adapt it to your own robot, cameras and tasks.

Some background

The biggest obstacle to teaching robots useful work is data. The usual approach is to have people remote-control robots through demonstrations over and over, which is slow and expensive. Runway argues that ordinary web video works as well as teleoperated robot video for pretraining, which shifts the bottleneck from how much robot data you have to how much video a model can learn from. That's why a video-generation company thinks it can compete in robotics.

Caveats

A 46.2% success rate leads the leaderboard, but it also means the model still fails more than half of the benchmark tasks, far from the reliability factories or homes would need. Every result comes from Runway or its partners and hasn't yet been independently verified.

What it means for you

It won't be in your home anytime soon. But a leaderboard-topping robot "brain" that's free and licensed for commercial use lets small robotics startups, university labs and even individual developers skip the huge cost of training from scratch. That could speed up how quickly robots show up in warehouses, factories and stores, and the robots you eventually run into sorting packages or restocking shelves may be running open models like this one.

Original sources published: Runway on X, Nov. 11, 2026, 23:04 Beijing time (UTC+8); Runway's website announcement is dated Nov. 11, 2026 (U.S. time), with no clock time given