NVIDIA expands Cosmos with data blueprints, training tools and Cosmos 3 models

Five blueprints for training data

World models help physical AI systems understand and simulate the real world, and NVIDIA says building them starts with multimodal, multisensory data. Cosmos Data Factory bundles tools and skills for labeling, curating, augmenting and evaluating that data at scale. Its Cosmos Curator component, a GPU-accelerated solution for curating video training datasets, now supports multimodal multisensory inputs and processing blocks.

Data Factory now offers five cloud-native blueprints, available through GitHub, Microsoft Azure, Nebius and CoreWeave:

• Autonomous Vehicle Dataset Generation: curates and labels driving data, then adds synthetic examples of hard-to-capture scenarios.
• Image Attribute Augmentation: varies object appearance to improve image search.
• Event Video Generation: turns still images into synthetic videos of rare or unsafe events.
• Defect Image Generation: creates printed circuit board, metal and glass defect images for inspection models.
• Video Data Augmentation: changes weather and lighting in existing footage while preserving scene structure.

NVIDIA named several early users. Porsche is leading a shared autonomous vehicle data flywheel across Volkswagen Group brands, using Data Factory through Voxel51 on Microsoft Azure. This is expected to reduce repeated physical test drives by up to 90% and shorten development cycle times by up to 85%, according to NVIDIA. ANYbotics generated over 1,500 synthetic video clips to train industrial-inspection robots to recognize liquid spills and other changes between facility visits. Ultralytics retrained its YOLO model with real images and targeted synthetic defects, improving defect localization and recall.

Cosmos Framework and Cosmos 3 for training and adaptation

At GTC Berlin, NVIDIA extended Cosmos Framework training, optimization and deployment tools and recipes as agent skills to more open models, alongside Cosmos 3. Developers and agents can now adapt open models such as GR00T, Alpamayo, QWEN, Pi0.5 and other compatible Hugging Face models through fine-tuning and reinforcement learning. They can then optimize models with distillation and quantization and deploy them with high-performing inference. Cosmos Framework is described as an open, modular framework of model training and optimization tools and skills.

Robotics research lab MicroAGI post-trained Cosmos 3 on videos of humans manipulating objects to build a world model specialized for egocentric manipulation. It distilled its 32-billion-parameter reasoner down to an 8-billion-parameter Nano model with no loss in accuracy, according to the source. Separately, developers can label data and post-train Cosmos 3 on Roboflow, then deploy to Roboflow hosted or on-premises NVIDIA GPUs to bring video reasoning into factories and warehouses.

Simulation models and faster inference

New Cosmos 3 Nano model variants for simulation let developers test autonomous policies in repeatable, hard-to-capture, multiview multisensor scenarios. Accompanying code allows them to adapt the models to train and evaluate their own policies for autonomous vehicles and bimanual robots in interactive simulations before real-world testing.

For deployment, updated NVIDIA Cosmos 3 NIM microservices deliver up to 2x faster inference with no quality loss in validated configurations, NVIDIA says, while continuous video analysis enables sub-second alerts. The new NVIDIA Metropolis Blueprint for video search and summarization (VSS) 3.3, built on Cosmos NIM microservices, turns one prompt into a video AI application and helps teams monitor 40% more video streams with the same GPU, according to the company.

ADNOC uses Fogsphere's MistIQ agentic platform, built with Cosmos and the Metropolis VSS Blueprint, to detect safety hazards in real time across 65 sites. Fogsphere also uses Cosmos Data Factory to generate synthetic training data for rare events, improving detection without physically staging dangerous situations.

What it means for developers

NVIDIA positions Cosmos as a fully modular and open platform covering the whole workflow: Data Factory to generate training datasets, Cosmos Framework to train and adapt world models, and NIM microservices to deploy supported Cosmos models. For teams working on autonomous vehicles, robots and vision AI, the release adds ready-made blueprints and simulation models aimed at reducing the need for physical testing and staged scenarios. The performance and savings figures come from NVIDIA and its partners, and the source does not include independent verification.