NVIDIA releases new open models to support autonomous and healthcare AI applications

NVIDIA expanded its open model families on Monday, targeting the development of autonomous agents and specialized applications in healthcare and physical AI. The expansion provides foundational models intended to power agentic systems capable of operating across diverse environments, from local workstations to cloud-based supercomputers.

The models are supported by the newly introduced NemoClaw stack, which enables single-command installation of NVIDIA Nemotron models and the OpenShell runtime. This platform includes an isolated sandbox and a privacy router designed to manage data security across local and cloud-based model deployments.

For large-scale infrastructure, NVIDIA released the Vera Rubin DSX reference design and the Omniverse DSX Blueprint. These frameworks provide standardized designs for AI infrastructure and digital twin platforms, intended to streamline the construction of data centers optimized for generative AI workloads.

The company also introduced hardware platforms for space-based data processing. The Space-1 Vera Rubin Module provides 25 times the AI compute performance of the H100 for orbital inferencing. Partners such as Starcloud and Sophia Space are utilizing these platforms to build purpose-designed orbital data centers and hosted computing infrastructure.

Supported hardware for the new software stacks includes GeForce RTX PCs, RTX PRO workstations, and DGX AI supercomputers. The integration of open models with these hardware platforms is intended to provide a consistent software environment for developers building autonomous agents.