NVIDIA Cosmos Opens New Tools to Train, Evaluate and Deploy World Models

The release adds five cloud-native Cosmos Data Factory blueprints, Cosmos 3 Nano simulation model variants, and updated Cosmos 3 NIM microservices offering up to 2x faster inference.
Key points
- Five new blueprints cover autonomous vehicle dataset generation, image attribute augmentation, event video generation, defect image generation, and video data augmentation
- Cosmos Curator adds support for multimodal multisensory inputs and GPU-accelerated processing blocks
- Cosmos Framework tools now support fine-tuning, reinforcement learning, and distillation on open models including GR00T, Alpamayo, Qwen, and Pi0.5
- New Metropolis Blueprint for video search and summarization 3.3 allows monitoring 40% more video streams per GPU
- Cosmos Data Factory blueprints are accessible via GitHub, Microsoft Azure, Nebius, and CoreWeave
Why it matters
Robotics and autonomous vehicle developers gain synthetic data pipelines and simulation checkpoints to train and evaluate physical AI systems before deploying to hardware.
Sources
- NVIDIA Cosmos Opens New Tools to Train, Evaluate and Deploy World Models (NVIDIA AI, Oct 23, 2026, 2:06 AM)