Open Source2026-06-02Hugging Face Blog

NVIDIA Cosmos 3: Open Omni-Model for Physical AI

NVIDIA has officially released Cosmos 3, a groundbreaking open omni-model designed specifically for physical AI reasoning and action. Unlike traditional AI models that operate solely in digital spaces, Cosmos 3 is built to help robots and autonomous systems understand, navigate, and interact with the physical world in real time. This release marks a significant step forward in embodied AI, where machines must bridge the gap between simulation and reality. Cosmos 3 processes multimodal inputs — including vision, touch, and spatial data — to generate coherent actions in dynamic environments. For example, a warehouse robot using Cosmos 3 can not only identify objects but also predict how they will behave when moved, allowing for safer and more efficient operations. The model is open-source, which NVIDIA hopes will accelerate research and development across the robotics and autonomous systems industries. By providing a foundational model that others can build upon, NVIDIA aims to democratize access to cutting-edge physical AI capabilities. Early applications are expected in manufacturing, logistics, healthcare robotics, and autonomous vehicles. The ability to reason about physics — such as gravity, friction, and object permanence — gives Cosmos 3 an edge over conventional AI models that lack real-world grounding. Industry analysts have praised the move, noting that physical AI has long been hampered by the lack of robust, open models. With Cosmos 3, NVIDIA is positioning itself at the center of the next wave of AI innovation, where machines don't just think — they act.

Related news