Nvidia released Cosmos 3, an open omni-model designed for physical AI reasoning and action. It integrates vision and language to help robots navigate and interact with real-world environments. This release provides a standardized framework for training embodied agents. Developers can now deploy these capabilities to improve robot precision and spatial awareness in complex settings.