Nvidia has released the Cosmos family of open models for world simulation

Nvidia has introduced the Cosmos family of open models designed for interactive physical world simulation based on multimodal data. The new models are developed for integration into robotics, autonomous transport, and computer vision systems.

The platform includes pre-trained neural networks for generating high-definition video, predicting the dynamics of physical objects, and modeling spatial interactions. The release aims to accelerate the development of autonomous systems and the training of robots in virtual environments.

The entire model stack and accompanying materials have been published on the Hugging Face platform under an open license. Developers have been granted access to model weights for deployment on their own servers and integration into workflows.

The release of the Cosmos suite allows research teams and engineers to test complex scenarios of AI interaction with physical reality without the need to collect expensive real-world data.