Nvidia Unveils Cosmos 3 AI World Model
Nvidia launched Cosmos 3, an open AI world model, to improve physical systems' real-world understanding and prediction.
Atlas Newsdesk ·
Nvidia Unveils Cosmos 3 AI World Model
Nvidia introduced Cosmos 3, an open artificial intelligence (AI) world model, on June 1, 2026, designed to enhance the understanding and predictive capabilities of robots, autonomous vehicles, and other physical systems in real-world environments. This development positions Nvidia to expand its influence beyond chip manufacturing into foundational AI models and software for physical AI applications.
Cosmos 3 was trained on 20 trillion tokens of multimodal data, including nearly one billion images, 400 million real and synthetic videos, ambient audio, text, and action data from humans and robots. This action data differentiates Cosmos 3 by modeling machine movement, enabling developers to simulate actions in physical environments and build task-specific models. The model generates action data, such as robot joint angles and gripper positions, to facilitate machine training for navigation and manipulation.
Nvidia is releasing two versions: a "super" model for high-physics accuracy tasks like robot and autonomous vehicle training, and a "nano" model for rapid results. An "edge" model for local operation is forthcoming. Cosmos 3 is an open model, allowing hardware manufacturers to customize it and ensuring alignment with industry needs. Initial partners supporting this effort include Agile Robots, Black Forest Labs, and Runway. The model can generate rare or dangerous scenarios, which are difficult or unsafe to capture in real-world settings.