NVIDIA has announced an expansion of its Cosmos platform, which is built around world models — AI systems designed to understand and simulate real physical environments. The update is intended to help developers working on robotics, autonomous vehicles and other physical AI applications that require an accurate representation of the real world.
The new additions include extra training tools, preconfigured simulation checkpoints, and blueprints for Cosmos Data Factory, a component focused on generating and managing synthetic data. Together, these resources are meant to let teams build, evaluate and deploy world models more quickly, without having to start each project from scratch.
NVIDIA highlights Cosmos's standing among open platforms in this space, noting it ranks in the top five across six different open model leaderboards. Such rankings should be read with some caution, since benchmark methodologies differ and companies naturally tend to promote results that favor their own products.
This expansion fits into NVIDIA's broader push to establish itself as a central player in physical AI, a fast-growing segment where the company hopes to replicate the dominance it built around GPUs for large language model training. By opening up more tooling around Cosmos, NVIDIA is also working to build a developer community that is likely to rely on its hardware and cloud services for training and deploying these models going forward.