NVIDIA Launches Advanced Open Models to Power Agentic, Physical and Healthcare AI

The Next Wave of Intelligent Agents

NVIDIA has launched a major expansion of its open model families during the GTC 2026 conference. These new tools allow developers to build intelligent agents that can reason and act across both digital environments and the physical world. The portfolio includes significant updates to the Nemotron, Cosmos, Isaac, Alpamayo, and BioNeMo platforms.

Multimodal Intelligence for Enterprise

The Nemotron 3 family introduces omni-understanding models that seamlessly integrate language, vision, and voice. Nemotron 3 Ultra offers five times the throughput efficiency on the NVIDIA Blackwell platform, making it ideal for complex tasks like coding assistants and workflow automation. Meanwhile, the new VoiceChat model enables real-time conversations where the AI can listen and respond simultaneously in a single, unified system.

Reasoning in the Physical World

For robotics and autonomous systems, NVIDIA unveiled Cosmos 3, a world foundation model designed to unify synthetic world generation with action simulation. The Isaac GR00T N1.7 model has reached commercial viability for humanoid robot deployment, while Alpamayo 1.5 supercharges autonomous vehicle navigation. These models allow machines to perceive and interact with their surroundings with unprecedented accuracy and safety.

Breakthroughs in Life Sciences

The BioNeMo platform is expanding its reach into drug discovery with Proteina-Complexa, a generative model for protein binder design. NVIDIA also collaborated with Google DeepMind and other partners to add millions of protein complex predictions to the AlphaFold database. Additionally, the new nvQSP engine accelerates medical treatment simulations by up to 77 times compared to traditional CPU-based methods.

Global Adoption and Availability

Industry leaders such as ServiceNow, CrowdStrike, and Novo Nordisk are already adopting these models to power their respective AI applications. These tools are being released as NVIDIA NIM microservices, ensuring they can be deployed securely across cloud and edge infrastructure. Developers can access the latest open models and datasets through build.nvidia.com and other leading AI platforms today.