- Runway has released Praxis-1, a robot-control model, with open weights and source code under the Apache 2.0 license.
- Praxis-1 uses knowledge from video training to predict scene changes and calculate matching robot movements, per Runway’s research report.
- On Nvidia’s RoboLab-120 leaderboard, Praxis-1 scores a 46.2 percent success rate, ahead of FLUX 3 Action’s 42.9 percent among World Action Models.
- Robotics companies Standard Bots, Noble Machines, and Ultra have tested Praxis-1 since September on humanoid, industrial, and warehouse robots.
What Happened
Runway has released Praxis-1, its robot-control model, with open weights and source code under the Apache 2.0 license, The Decoder reported on November 14, 2026. The company first introduced Praxis-1 in late September but had limited access to select partners before this wider release.
Why It Matters
Runway, known for generative video models including Gen-4.5, is extending its video research into robotics as general-purpose models increasingly compete with specialized robotics systems. The release follows Agility Robotics’ debut as the first humanoid robot maker to go public, underscoring how fast the field is attracting new entrants.
Technical Details
Runway calls Praxis-1 a “World Action Model” that predicts how a scene will change and derives matching robot movements, using camera images, text instructions, and position data as input. Runway fine-tuned the underlying video model using roughly 277 hours of the public DROID dataset for a Franka arm and between 1,400 and 2,600 hours of training material for a dual-arm robot. On Nvidia’s RoboLab-120 leaderboard, which tests 120 sorting, grasping, and stacking tasks, Praxis-1 scored 46.2 percent, ahead of FLUX 3 Action’s 42.9 percent among World Action Models.
Who’s Affected
Robotics companies Standard Bots, Noble Machines, and Ultra have been testing Praxis-1 since September on a humanoid robot, an industrial arm, and a dual-arm warehouse robot, giving them an open-weight alternative to proprietary robot-control systems.
What’s Next
Runway acknowledges Praxis-1 still occasionally confuses similar-looking objects and cannot yet reliably plan complex action sequences on its own, leaving open how much robot-specific data the approach ultimately needs compared with models that skip video pretraining.