Nvidia is expanding its open artificial intelligence ecosystem with new models and a global collaboration to build next-generation AI systems. The announcements came at Nvidia’s Global Technology Conference in San Jose.
The company launched the Nemotron 3 family of multimodal models, including Nemotron 3 Ultra, which delivers five times greater throughput efficiency using Nvidia’s NVFP4 format on Blackwell GPUs. Nemotron 3 Omni integrates audio, vision, and language, while Nemotron 3 VoiceChat enables real-time conversational AI.
Safety enhancements include Nemotron safety models and an agentic retrieval pipeline to reduce hallucinations and improve accuracy.
Enterprise adopters include Automation Anywhere, CrowdStrike, Perplexity AI, ServiceNow, and LangChain. In robotics, Nvidia introduced Cosmos 3 for simulation, Isaac GR00T N1.7 for humanoid robots, and Alpamayo 1.5 for autonomous vehicles.
In life sciences, BioNeMo now features Proteina-Complexa, a generative model accelerating drug discovery by designing protein binders. Nvidia partnered with Google DeepMind, EMBL-EBI, and Seoul National University to expand the AlphaFold database with millions of predicted protein complexes.
The new Nemotron Coalition, co-founded with Mistral AI, aims to co-develop open frontier models using shared data, expertise, and DGX Cloud computing. The first joint model will underpin the upcoming Nemotron 4 series.