NVIDIA Cosmos Platform Released for Building World Models in Physical AI
NVIDIA introduces Cosmos, a new platform offering foundation world models and development tools aimed at physical AI developers. The Cosmos Framework provides agent skills and recipes to support world model training and deployment.
NVIDIA has launched Cosmos, a platform that supplies foundation world models and a suite of development tools specifically for physical AI applications. The release includes the Cosmos Framework, which offers agent skills and training recipes to help developers build and deploy world models.
What changed?
NVIDIA Cosmos is now available, delivering a platform that combines world foundation models with a development framework. The Cosmos Framework provides tools, agent skills, and training recipes tailored for world model creation and adaptation in physical AI contexts.
Why does it matter to an everyday developer?
Physical AI developers often struggle to adapt world models to their sensors, tasks, and data, as well as reliably deploy those models. Cosmos addresses these pain points by offering structured tools and recipes, potentially reducing development friction and improving model reliability. This can streamline workflows for developers in robotics, simulation, or interactive environments who need adaptable and robust world modeling capabilities.
What can the developer do now?
Developers can evaluate Cosmos for their physical AI projects, using the provided foundation models as starting points and the Cosmos Framework to access agent skills and model training recipes. This allows faster iteration for building, customizing, and deploying world models. Those working with complex sensor setups or interactive environments are especially likely to benefit from the new tools.
