Nvidia launches Cosmos 3 as an open physical-AI foundation model

Cosmos 3 is a missed-but-still-important physical-AI item from the latest sweep. Nvidia says the model family is a fully open omnimodel for robotics, autonomous vehicles, industrial vision, and synthetic data generation, with native understanding and generation across text, image, video, ambient sound, and action. The company says Cosmos 3 uses a mixture-of-transformers architecture that separates reasoning and generation, and that it leads among open models on several physical-AI, world-generation, action-policy, and vision-understanding benchmarks. Nvidia also launched a Cosmos Coalition with Agile Robots, Black Forest Labs, Generalist, LTX, Runway, and Skild AI, and says Cosmos 3 Super and Nano are available now while Cosmos 3 Edge is coming later. The strategic point is that Nvidia is packaging open models, synthetic data, cloud training, and robot tooling as one physical-AI platform.

Key details: Nvidia, May 31, 2026, Cosmos 3, open physical AI foundation model, mixture-of-transformers architecture, text image video sound and action, Cosmos Coalition, Agile Robots.

Continue swiping for more AI Brief stories.