Cosmos 3

Cosmos 3 is a family of omnimodal world models for Physical AI. The models jointly process and generate language, images, video, audio, and action sequences using a unified Mixture-of-Transformers architecture. Cosmos 3 supports Reasoner and Generator runtime surfaces for world understanding, world generation, and action modeling.

For model code and examples, see the Cosmos GitHub repository. Model artifacts are available in the NVIDIA Cosmos 3 Hugging Face collection. For technical details, see the Cosmos 3 Technical Report.

Training Content

Review the Cosmos 3 training content summary.