NVIDIA Expands Cosmos Platform With New World Model Training and Deployment Tools

NVIDIA expands Cosmos with new data, training and deployment tools

NVIDIA has added new training tools, simulation checkpoints and Cosmos Data Factory blueprints to its Cosmos platform, the company said, aiming to help developers build, evaluate and deploy world models for physical AI systems. NVIDIA says the Cosmos platform ranks in the top five on six different open models leaderboards.

Cosmos Data Factory blueprints

NVIDIA says Cosmos Data Factory brings together tools for labeling, curating, augmenting and evaluating data at scale for training physical AI, and that its Cosmos Curator tool now supports multimodal, multisensory inputs and processing blocks.

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

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

NVIDIA says Porsche is using Cosmos Data Factory through Voxel51 on Microsoft Azure to build a shared autonomous vehicle data flywheel across Volkswagen Group brands, which the company says is expected to reduce repeated physical test drives by up to 90% and shorten development cycle times by up to 85%. NVIDIA also says ANYbotics generated more than 1,500 synthetic video clips using the blueprints, validated with Encord's platform, to train inspection robots to recognize liquid spills, and that Ultralytics retrained its YOLO model with synthetic defects generated through Cosmos Data Factory to improve defect localization and recall.

Training, adapting and optimizing models

NVIDIA says it extended Cosmos Framework training, optimization and deployment tools to more open models at GTC Berlin, alongside a new model called Cosmos 3. According to the company, developers can now adapt open models including GR00T, Alpamayo, QWEN and Pi0.5, as well as other compatible Hugging Face models, through fine-tuning and reinforcement learning, then optimize them with distillation and quantization before deployment.

NVIDIA says robotics research lab MicroAGI post-trained Cosmos 3 on videos of humans manipulating objects and distilled its 32-billion-parameter reasoner into an 8-billion-parameter Nano model with no loss in accuracy. The company also says developers can label data and post-train Cosmos 3 on Roboflow, then deploy the resulting models to Roboflow-hosted or on-premises NVIDIA GPUs.

Testing policies and running models faster

NVIDIA says new Cosmos 3 Nano model variants for simulation let developers test autonomous policies in repeatable, multiview, multisensor scenarios before real-world testing. The company says updated Cosmos 3 NIM microservices deliver up to 2x faster inference with no quality loss in validated configurations, and that continuous video analysis built on these microservices enables sub-second alerts.

NVIDIA says its Metropolis Blueprint for video search and summarization, version 3.3, is built on Cosmos NIM microservices and helps teams monitor 40% more video streams using the same GPU. The company says ADNOC uses Fogsphere's MistIQ platform, built with Cosmos and the Metropolis VSS Blueprint, to detect safety hazards in real time across 65 sites, and uses Cosmos Data Factory to generate synthetic training data for rare events.

Source: NVIDIA, "NVIDIA Cosmos Opens New Tools to Train, Evaluate and Deploy World Models," published October 21, 2026.