NVIDIA Cosmos 3 is the world’s first fully open omni-model for physical AI, combining vision reasoning, world generation, and action prediction in a single foundation model announced May 31, 2026.
Among open models, Cosmos 3 ranks first on more than eight leaderboards for vision reasoning, text-to-image, image-to-world, and world-action generation with state-of-the-art physics accuracy.
Built on mixture-of-transformers architecture pairing autoregressive reasoner with diffusion-based generator. Cosmos 3 Nano and Super checkpoints, open code, post-training recipes, and datasets for robotics, autonomous driving, physics simulation, human motion, and warehouse operations released.
Cosmos Coalition: Agile Robots, Black Forest Labs, Generalist, LTX, Runway, Skild AI.