Home
Latest reporting
News
NVIDIA added training tools, simulation checkpoints and data blueprints to Cosmos to support physical AI development and deployment.
NVIDIA said it has expanded its Cosmos platform with new training tools, simulation checkpoints and Cosmos Data Factory blueprints as it looks to speed the development of world models for physical AI systems. The company said the additions are designed to help developers build training data, adapt models, test behavior and deploy systems with fewer manual steps.
Evidence and context
In a blog post published Oct. 22, 2026, NVIDIA said Cosmos Data Factory now includes five cloud-native blueprints for tasks including autonomous vehicle dataset generation, image attribute augmentation, event video generation, defect image generation and video data augmentation. The company said the tools are available through GitHub, Microsoft Azure, Nebius and CoreWeave. NVIDIA also said Cosmos Curator has added support for multimodal multisensory inputs and processing blocks.
The company cited several early users of the blueprints. NVIDIA said Porsche is using Cosmos Data Factory through Voxel51 on Microsoft Azure to build a shared autonomous-vehicle data flywheel across Volkswagen Group brands, with an expected reduction in repeated physical test drives of up to 90% and development cycle times by up to 85%. NVIDIA said ANYbotics generated more than 1,500 synthetic video clips with Cosmos Data Factory blueprints to train industrial inspection robots, while Ultralytics retrained its YOLO model with real images and synthetic defect examples to improve defect localization and recall.
NVIDIA also said it has extended Cosmos Framework training, optimization and deployment tools as agent skills to more open models, including Cosmos 3, at GTC Berlin. According to the company, developers can fine-tune and reinforce models such as GR00T, Alpamayo, QWEN, Pi0.5 and other compatible Hugging Face models, then apply distillation and quantization before deployment. The company said MicroAGI post-trained Cosmos 3 on videos of human object manipulation and distilled its 32-billion-parameter reasoner into an 8-billion-parameter Nano model with no loss in accuracy.
For simulation and evaluation, NVIDIA said new Cosmos 3 Nano variants support repeatable testing of autonomous policies in multiview, multisensor scenarios, including for autonomous vehicles and bimanual robots. It also said updated Cosmos 3 NIM microservices can deliver up to 2x faster inference with no quality loss in validated configurations, while continuous video analysis enables sub-second alerts. NVIDIA said its new Metropolis Blueprint for video search and summarization 3.3, built on Cosmos NIM microservices, can help teams monitor 40% more video streams with the same GPU.
The company framed Cosmos as a modular platform for generating training datasets, training and adapting world models, and deploying supported models through NIM microservices. Background material verified on Oct. 22, 2026, places Cosmos within NVIDIA’s broader physical AI stack for robots, autonomous vehicles and vision AI, while noting known limitations such as temporal inconsistency, object morphing and implausible dynamics that require additional validation in safety-critical uses.
