NVIDIA has introduced Cosmos 3, its latest open foundation model designed specifically for physical AI applications.
Built on a new mixture-of-transformers architecture, Cosmos 3 combines vision reasoning, world simulation, and action generation within a single system. NVIDIA describes it as the world's first fully open omnimodel capable of understanding and generating text, images, video, ambient sound, and actions.
The model is designed to help developers create robots, autonomous vehicles, and intelligent systems that can perceive, reason, plan, and act in real-world environments. NVIDIA says Cosmos 3 can significantly reduce training and evaluation cycles while accelerating synthetic data generation and physical AI development.
