Nvidia has expanded its open model lineup with tools for AI agents, structured business data, robotics, autonomous vehicles, protein design and quantum computing. The company says organizations can customize the models with their own data and run them on infrastructure they control.
The language lineup adds Nemotron 3.5 Super VL and Nemotron 3.5 Bolt, alongside Nemotron Retriever models for embedding, reranking and parsing documents. Nvidia says Nemotron supports agents that work with text and images, while Retriever helps them find and use information.
Kumo works with business data
Nvidia also introduced Kumo, a family of open models and libraries for predictions from structured business data. Its tools cover forecasting, anomaly detection and tasks such as fraud risk scoring. Developers can use pretrained models or fine-tune them with proprietary data, according to the company. Cognite, Coinbase, Databricks, DoorDash, Revolut and Tubi are among the organizations working with Kumo.
Cosmos adds simulation and data tools
For robotics and driving systems, Nvidia introduced Cosmos 3.5 world models, tools for refining world models with less compute, and simulation checkpoints for evaluating behavior before real-world tests. Cosmos Data Factory adds workflows for curating, labeling, generating and evaluating training data.
Alpamayo 2 Edge is a compact model designed to understand driving scenes and plan vehicle motion in real time. Nvidia says it is optimized for the DRIVE AGX Thor platform and requires less computing capacity to bring those capabilities into vehicles.
In science, Nvidia and the Baker Lab at the University of Washington’s Institute for Protein Design released BioNeMo RFD4-Proteina, which generates protein sequences and 3D structures. Nvidia says it generated designs about 13 times faster than RFdiffusion3 in a benchmark involving proteins of 200 amino acids. Researchers can evaluate its proposed designs before choosing candidates for laboratory tests.
Read nextNvidia expands Isaac robotics platform as Agile Robots unveils reference humanoidNvidia’s Ising model family targets quantum hardware calibration and error correction. IQM Quantum Computers is using it to calibrate processors, while AWS is integrating Ising into AWS Parallel Computing Service so researchers can work with the models before deployment.
Nemotron 3.5 Super VL, Retriever, Kumo, Cosmos, RFD4-Proteina and Ising are available through the platforms Nvidia lists for each release. Nemotron 3.5 Bolt E4B 4GB is expected on October 23, and Alpamayo 2 Edge is due on GitHub and Hugging Face later this month.
