NVIDIA expands open models for locally controlled AI

NVIDIA is expanding its open model portfolio to help organizations customize AI with local data and deploy it on infrastructure they control. The additions span language, structured business data, physical simulation, autonomous driving, biology and quantum computing, supporting applications tailored to regional languages and specialized industries.

New Nemotron 3.5 Super VL and Nemotron 3.5 Bolt models support enterprise agents, alongside Nemotron Retriever models for embedding, reranking and document parsing. Companies including Dassault Systèmes, SAP and Salesforce are working with Nemotron models and datasets. NVIDIA Kumo adds models and libraries for forecasting, fraud risk scoring and recommendations using structured business data, with early users including Coinbase, Databricks and Revolut.

Cosmos gains post-training tools, simulation checkpoints and Data Factory workflows for curating, generating and evaluating training data. Porsche, Siemens and other companies are using Data Factory to develop datasets. Alpamayo 2 Edge brings scene understanding and motion planning into autonomous vehicles, with lower computing requirements. It is expected on GitHub and Hugging Face later this month.

BioNeMo RFD4-Proteina, developed with the University of Washington’s Baker Lab, generates protein sequences and structures for research. NVIDIA reports designs about 13x faster than RFdiffusion3 in a benchmark of 200 amino acid proteins. Ising supports quantum hardware calibration and error correction, with adoption by IQM Quantum Computers and SEEQC and integration into AWS Parallel Computing Service.