Runway open-sources Praxis-1, a 27B MoE video-action model that ranks #1 on RoboLab-120
runwayml · x · 2026-11-13
- What it is: Runway's Praxis-1 is an open-weight world action model built from a 27B-parameter mixture-of-experts diffusion transformer that jointly predicts video and actions, turning the company's video pretraining into real robot control.
- Results: Ranks #1 among world action models and all open-source robot foundation models on NVIDIA's RoboLab-120 leaderboard; competitive success rates on real-world Franka tasks vs. state-of-the-art policy models.
- Capabilities: Controls a bimanual YAM robot and a single Franka arm on contact-rich everyday tasks (upside-down mug placement, box packing, bottle cleanup); adapts quickly to new robots and can be fine-tuned on a low-cost SO-101.
- Deployment: Built on thousands of LeRobot community datasets, runs on a single RTX 5090 with NVFP4; checkpoint and setup instructions are available.
Related event: Runway Open-Sources Praxis-1 World Action Model, Tops RoboLab-120(20 posts)→
More from Embodied
- X-Era AI Lab's world-action model hits 85-88% robot manipulation success rates — jiqizhixin · 2026-11-14
- Figure AI seen most exposed to Nvidia fallout: $3.5-6B compute deal in limbo — kscottz · 2026-11-14
- 5,400 real-robot rollouts benchmark Claude Fable 5.2 at 65%, ahead of Gemini and GPT — ricklamers · 2026-11-14
- EmbodiedSWE wins CoRL 2026 Best Paper: coding agents generate robot training data — PeterHndrsn · 2026-11-14
- Ducted-Fan Robot Uses Coandă Effect for 3D Thrust Vectoring and Stable Flight — bennash · 2026-11-14
- ASUS teases its first humanoid robot in near-black teaser video — Distinct-Question-16 · 2026-11-13