Runway releases open-weight robotics model Praxis-1
Runway released Praxis-1, a 27-billion-parameter open-weight model for predicting robot video and actions, under the permissive Apache 2.0 license. The company claims it tops NVIDIA's RoboLab-120 leaderboard with a 46.2% success rate across 1,200 episodes, and it supports full and LoRA fine-tuning across different robot hardware.
Why it matters
๐ป Developer ยท Open weights on GitHub and Hugging Face with LoRA fine-tuning support make this immediately usable for robotics experimentation without starting from scratch.
๐ฆ Product ยท A credible open alternative to closed robotics foundation models lowers the cost of building robot-control products.
๐จ Design ยท Robotics video-action prediction has UX implications for how operators supervise and correct robot behavior in real time.
๐ Business ยท Open licensing under Apache 2.0 means companies can build commercial robotics products on this without royalty or usage restrictions.
๐ค Just Curious ยท A video-game-adjacent AI company just open-sourced a model that helps robots figure out what to do next, and claims it beats the competition.
Sources: Runway Praxis-1 Is Now Available: Open-Weight Robotics Model