Welcome NVIDIA Cosmos 3: The First Open Omni-model for Physical AI Reasoning and Action

NVIDIA Cosmos 3 is a unified omni-model for physical AI that combines world generation, physical reasoning, and action generation into a single system. Built on a Mixture-of-Transformers (MoT) architecture, it processes text, image, video, and action inputs to generate realistic and physically plausible video simulations. The model is designed for applications in robotics, autonomous vehicles, and smart spaces by enabling systems to understand motion, causality, and physics. It is released in two sizes, a 16B parameter Nano model and a 64B parameter Super model, which are available on Hugging Face.