The updated platform, announced this week, ranks among the top five on six different open models leaderboards and provides world model builders with frontier open models along with data, training and optimization tools and recipes.
NVIDIA Cosmos Data Factory, a central component of the expansion, combines tools and skills for labeling, curating, augmenting and evaluating data at scale. The GPU-accelerated Cosmos Curator now supports multimodal multisensory inputs and processing blocks. Five cloud-native blueprints are available through GitHub, Microsoft Azure, Nebius and CoreWeave, covering autonomous vehicle dataset generation, image attribute augmentation, event video generation, defect image generation and video data augmentation.
Companies are already using the blueprints. Porsche is leading development of a shared autonomous vehicle data flywheel across Volkswagen Group brands using Cosmos Data Factory through Voxel51 on Microsoft Azure. NVIDIA stated this is expected to reduce repeated physical test drives by up to 90% and shorten development cycle times by up to 85%.
ANYbotics generated over 1,500 synthetic video clips with the blueprints, curating real-world data and validating outputs with Encord’s platform to train industrial-inspection robots to recognize liquid spills and other changes. Ultralytics, a computer vision company, retrained its YOLO model using Cosmos Data Factory to generate targeted synthetic defects, improving defect localization and recall.
On the training side, NVIDIA extended Cosmos Framework with agent skills to more open models, including GR00T, Alpamayo, QWEN and Pi0.5. Developers can adapt models through fine-tuning and reinforcement learning, then optimize via distillation and quantization before deployment. Robotics research lab MicroAGI post-trained Cosmos 3 on videos of human object manipulation and distilled a 32-billion-parameter reasoner down to a 8-billion-parameter Nano model without accuracy loss. Models can also be labeled and post-trained on Roboflow for specific environments and tasks, then deployed on NVIDIA GPUs.
For testing and deployment, new Cosmos 3 Nano model variants enable simulation of autonomous policies in repeatable, multiview multisensor scenarios. Updated NVIDIA Cosmos 3 NIM microservices deliver up to 2x faster inference with no quality loss. The NVIDIA Metropolis Blueprint for video search and summarization 3.3, built on Cosmos NIM microservices, helps teams monitor 40% more video streams with the same GPU.
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. For rare events, Fogsphere also uses Cosmos Data Factory to generate synthetic training data, improving detection without staging dangerous situations.
NVIDIA described the Cosmos platform as fully modular and open, allowing developers to use the data factory, framework and NIM microservices as needed.