NVIDIA Announces Cosmos 3, an Open, State-of-the-Art Foundation Model for Physical AI

NVIDIA today announced NVIDIA Cosmos 3, an open world foundation model for physical AI based on a groundbreaking mixture-of-transformers architecture that integrates vision reasoning, world generation, and action prediction into a single system. Cosmos 3 is a fully open omni-model capable of natively understanding and generating text, images, video, environmental sounds, and actions. NVIDIA also launched the Cosmos Coalition, a global ecosystem initiative including partners like Agile Robots and Runway, to advance next-generation world models. CEO Jensen Huang stated that a 'Big Bang' of physical AI is approaching, enabling developers to make generational leaps in building robots and autonomous vehicles. Cosmos 3 addresses the fundamental challenge of generalization in the real world with limited training data and has achieved top rankings on major benchmarks. Cosmos 3 Super and Nano are available now, with the Edge version coming soon.