NVIDIA expands Cosmos tools for training and deploying world models
🎧 Listen to this story
NVIDIA announced at GTC Berlin an expansion of Cosmos, its platform for developing world models for physical AI. The additions include tools for labeling, curating, augmenting and evaluating multimodal data, as well as five Cosmos Data Factory blueprints for generating and processing synthetic data. They are available through GitHub, Microsoft Azure, Nebius and CoreWeave.
NVIDIA also extended Cosmos Framework’s training, optimization and deployment tools to compatible open models on Hugging Face. Developers can adapt models such as GR00T, Alpamayo, QWEN and Pi0.5 through fine-tuning and reinforcement learning, and use distillation and quantization. NVIDIA says Porsche expects the technology to reduce repeated physical test drives by up to 90% and shorten development cycles by up to 85%; ANYbotics generated more than 1,500 synthetic video clips to train inspection robots.
Why it matters · editorial interpretation
For teams developing physical AI, the expansion brings critical data, training, optimization and deployment steps into one offering while broadening options for adapting open models. Availability through multiple providers could make these workflows easier to adopt, although the cited gains remain tied to the expectations of the companies mentioned.