NVIDIA Unveils Multimodal AI Models at GTC 2026, Focusing on Agents, Robotics, and Healthcare Research

At the 2026 GTC conference, NVIDIA announced a comprehensive expansion of its open-source AI model family to accelerate advancements in AI agents, physical AI, and healthcare research. The newly launched Nemotron-3 series includes three models—Ultra, Omni, and VoiceChat—optimized respectively for code assistance, multimodal processing, and real-time conversation, with Nemotron-3 Ultra delivering a 5x improvement in throughput efficiency. In physical AI, NVIDIA introduced Cosmos-3, a world foundation model, and unveiled Isaac GR00T N1.7 and Alpamayo 1.5—vision-language-action models designed for humanoid robots and autonomous driving—both now commercially available. The next-generation GR00T N2 is expected to launch by year-end, doubling task success rates. In healthcare, the BioNeMo platform now features the Proteina-Complexa generative model. In collaboration with DeepMind and others, NVIDIA has added 1.7 million predicted protein complexes to the AlphaFold database. Additionally, NVIDIA launched nvQSP, a GPU-accelerated simulation engine that runs 77x faster than CPU-based alternatives. All models and tools are now available on GitHub and Hugging Face, and developers can rapidly deploy them via NVIDIA NIM microservices.