Runway open-sources 27B world action model Praxis-1, tops NVIDIA RoboLab-120
runwayml · x · 2026-11-11
Runway released Praxis-1, an open-weight (Apache 2.0) world action model that turns video pretraining into control for real robots.
- Architecture: 27B-parameter MoE diffusion transformer that jointly predicts video and actions; pretrained on general video to learn object dynamics, hand motion and task progression
- Results: #1 on NVIDIA's RoboLab-120 (46.2%) ahead of all open-weight VLAs including π0.5 (28.0%) and GR00T N1.6 (7.3%); 80.0% on SimplerEnv fine-tuned on RT-1 vs 52.4% for the original policy, with per-task wins like 98.3% on Pick coke can
- Adaptability: under 30 minutes of teleoperated demos sufficed for all 15 simulated trials on Noble Machines' robot; can also be fine-tuned on the low-cost SO-101
- Demos: controls bimanual YAM and Franka arms on contact-rich everyday tasks (flipping mugs, boxing items, clearing bottles)
Weights and technical report are public for running and fine-tuning on your own hardware.
Related event: Runway Open-Sources Praxis-1 World Action Model, Tops RoboLab-120(10 posts)→
More from Embodied
- Open-source dual-arm robot solves tangram puzzles with 90 demos, ~60% success — DanGrover · 2026-11-12
- Robotics has three regional playbooks: China volume, Europe inspection, US pilots — ingliguori · 2026-11-12
- 3,500 Laser-Armed AI Submarine Robots Patrol Norwegian Salmon Farms — philfung · 2026-11-12
- Model Autonomously Designs a Surgical-Robot-Style Robotic Arm CAD After 75 Minutes — _Stocko_ · 2026-11-12
- Dev releases local world model turning any image into interactive video, runs on RTX 3060 — lucidml_lover · 2026-11-12
- RoboRisk benchmark shows foundation models pose high harm rates in robotics; LEASH harness cuts risk — UoLiverpool · 2026-11-12