NVIDIA expands Cosmos platform with data blueprints, training tools and faster NIM microservices

NVIDIA has expanded its Cosmos platform for building world models with new training tools, simulation checkpoints and five cloud-native Cosmos Data Factory blueprints, available through GitHub, Microsoft Azure, Nebius and CoreWeave, according to an NVIDIA blog post.

One blueprint, Autonomous Vehicle Dataset Generation, curates and labels driving data and adds synthetic examples of hard-to-capture scenarios. NVIDIA says updated Cosmos 3 NIM microservices deliver up to 2x faster inference with no quality loss, but only in validated configurations, and that continuous video analysis enables sub-second alerts.

A later NVIDIA AI post on X says NVIDIA added Cosmos tools for generating training data, training world models and testing them before deployment. It names Porsche, Mercedes-Benz, Encord, Siemens, Fogsphere and LTX as Cosmos users across robotics, autonomous driving and factory operations, but does not describe their results.

A later NVIDIA Technical Blog post describes Cosmos as offering world foundation models and development tools for adapting models to sensors, tasks and data. The excerpt does not give benchmark scores, parameter counts, pricing or availability details.