NVIDIA Cosmos

NVIDIA Cosmos is an open physical AI foundation model built on a Mixture-of-Transformers architecture. It serves as an omni-model capable of native reasoning, world simulation, and action generation. Cosmos can be used as a vision language model (VLM) for reasoning over objects, interactions, and intent in complex real-world scenarios; as a backbone for world action models to accelerate robot policy learning; as a controllable, physics-grounded world simulator to predict and evaluate multiple approaches; and for scalable synthetic video data generation from text, image, video, ambient sound, and action inputs. The model is intended to accelerate the development of physical AI without being constrained by real-world data capture limitations.