At the annual GTC conference, NVIDIA announced a major expansion of its open model families designed to advance agentic, physical, and healthcare artificial intelligence. The initiative introduces new models and frameworks intended to enable developers and scientists to build intelligent systems capable of reasoning and acting across both digital and real-world environments. Kari Briski, NVIDIA's vice president of generative AI software, emphasized that open source AI has become a global innovation force, extending intelligence beyond simple language tasks to cover biology, robotics, and autonomous machines. The expanded portfolio includes the NVIDIA Nemotron 3 family, which powers agentic AI with omni-understanding capabilities across language, vision, voice, and safety. These models deliver natural conversations and complex reasoning, enabling enterprises to deploy specialized AI assistants. Major companies such as CodeRabbit, CrowdStrike, Perplexity, and ServiceNow are already utilizing Nemotron models to build advanced agentic applications. Additionally, LangChain has integrated these models into its platform to help businesses automate complex tasks at scale. To support local cultural alignment, NVIDIA released Nemotron-Personas, a collection of privacy-preserving synthetic datasets based on local census data for countries including the United States, Japan, India, Brazil, Singapore, and France. For physical AI, NVIDIA is accelerating autonomous systems with new foundation models. NVIDIA Cosmos 3 serves as the first world foundation model to unify synthetic world generation and physical AI reasoning, while the NVIDIA Isaac GR00T N1.7 is now commercially viable for humanoid robot deployment. The NVIDIA Alpamayo 1.5 model enhances autonomous vehicle reasoning with advanced navigation guidance and multi-camera support. During the keynote, CEO Jensen Huang previewed GR00T N2, a next-generation robot model expected to enable robots to succeed at new tasks in new environments more than twice as often as current models. This model is projected to rank number one for generalist robot policies and will be available by the end of the year. Industry leaders like LG Electronics, Toyota Research Institute, and Johnson & Johnson MedTech are actively adopting these tools. In the healthcare sector, the NVIDIA BioNeMo platform is expanding to accelerate biomedical research. A key addition is the Proteina-Complexa model, which aids in protein binder design and structure-based drug discovery. Companies including Novo Nordisk and Viva Biotech are using this model to design and test new protein therapeutics. Furthermore, NVIDIA collaborated with Google DeepMind and the European Bioinformatics Institute to massively expand the AlphaFold Protein Structure Database, adding millions of new protein complex predictions. The launch of nvQSP, a GPU-accelerated simulation engine, also marks a significant breakthrough, offering up to 77 times faster performance than traditional CPU simulations for exploring treatment scenarios. Most of these new open models, data sets, and frameworks are available on GitHub, Hugging Face, and the NVIDIA build platform. Many are also offered as NVIDIA NIM microservices for secure and scalable deployment across cloud and edge infrastructure. This expansion underscores NVIDIA's commitment to democratizing access to advanced AI tools, empowering a global community of developers to drive breakthroughs in diverse industries.