Nvidia unveiled its latest artificial intelligence platform, named Cosmos, during the Consumer Electronics Show (CES) 2025 on Monday. This innovative platform features a range of generative world foundation models (WFMs), as well as advanced tokenisers, an accelerated video processing pipeline, and various guardrails aimed at facilitating the development of physical AI systems, including autonomous vehicles and robotics. Furthermore, the company has made the WFMs available as open source for academic and research initiatives. Alongside this launch, Nvidia also presented the Llama Nemotron family of AI models at CES 2025.
Introducing Nvidia Cosmos World Foundation Models
In an official announcement, Nvidia outlined the capabilities of its new Cosmos platform. It is designed to host WFMs alongside tools that support the training and development of physical AI systems—machines equipped with mechanical components designed to interact with the real world.
The tech company emphasized that the creation of physical AI systems, such as robots and self-driving vehicles, often entails significant costs due to the need for extensive real-world data and a variety of testing environments. The Cosmos platform’s WFMs are engineered to address these challenges effectively.
Nvidia claims that these world AI models can produce large quantities of photorealistic and physics-based synthetic data, which can be utilized for training physical AI systems. Additionally, this data can aid in evaluating existing robots by exposing them to various testing scenarios. The Cosmos platform also enables developers to create tailored models through fine-tuning of the WFMs.
The AI models incorporated within Nvidia Cosmos provide video search and understanding capabilities, allowing developers to easily locate specific training videos within a vast database. By leveraging the Nvidia Omniverse platform, the Cosmos models can generate physics-based controlled 3D environments. The platform offers simulation-based training specifically designed for physical AI applications.
These AI models are available under an open model license, with access for developers through Nvidia’s application programming interface (API) catalog or via Hugging Face. Nvidia disclosed that numerous companies specializing in robotics and physical AI have already integrated the Cosmos platform into their operations, including 1X, Agile Robots, Agility, Figure AI, Foretellix, Fourier, Galbot, Hillbot, IntBot, Neura Robotics, Skild AI, Virtual Incision, Waabi, XPENG, and Uber.
