NVIDIA expands open models across AI, robotics, biology and quantum
NVIDIA said on Wednesday it is expanding its open-model portfolio across language, robotics, biology and quantum computing, adding new models and tools it says can be customized with local data and deployed on infrastructure customers control. The announcement broadens a family of releases that NVIDIA says is intended to help developers, researchers and enterprises build applications tailored to local languages and industries.
Evidence and context
In the Oct. 21, 2026 blog post, NVIDIA said the portfolio includes Nemotron 3.5 Super VL and Nemotron 3.5 Bolt; Nemotron Retriever models for embedding, reranking and document parsing; the Kumo family of open-weight foundation models and libraries for structured business data; Cosmos 3.5 world models; Alpamayo 2 Edge for autonomous vehicles; BioNeMo for healthcare and scientific discovery; and Ising for quantum computing. The company said the open models are designed to give organizations a foundation they can customize with local data and run on their own infrastructure.
NVIDIA tied the language-model side of the portfolio to agentic systems. It said Nemotron models support text and images and are intended to help agents orchestrate workflows, reason through complex problems and complete specialized tasks at lower cost. The company said the models were built with data, research and expertise from the NVIDIA Nemotron Coalition, including European partners Black Forest Labs, H-Company and Mistral. NVIDIA also said Dassault Systèmes, Domyn, Dream, Palantir’s customers, Salesforce, SAP, Shopify, Synopsys and WPP, along with Accenture and EY, are working with Nemotron models and datasets.
For structured enterprise data, NVIDIA said Kumo models and the Kumo Engine library are meant to help agents forecast outcomes, score fraud risk and recommend actions from relational and tabular data. The company said Kumo Structured includes models for prediction, forecasting and anomaly detection, and that Cognite, Coinbase, Databricks, DoorDash, Revolut and Tubi are among the first to work with the models. NVIDIA added that Snowflake customers are already using Kumo models in industries including advertising, energy, financial services and retail.
On physical AI, NVIDIA said Cosmos is an open world-model family for robotics, autonomous vehicles and vision agents. It said new post-training and optimization tools are designed to improve world models using less compute, while simulation checkpoints are intended to let developers evaluate behavior before real-world testing. NVIDIA also introduced Cosmos Data Factory, which it said combines blueprints and core agent skills for curating, labeling, generating and evaluating training data for driving scenarios, product defects, image augmentation and event generation. The company said Porsche with Voxel51, ANYbotics with Encord, Zalando and CEVA Logistics with Sereact, and Siemens are using Cosmos Data Factory.
NVIDIA said Alpamayo 2 Edge is a compact reasoning model for real-time scene understanding and motion planning inside autonomous vehicles, built on Cosmos 3 Edge’s Nemotron-based foundation and optimized for the NVIDIA DRIVE AGX Thor platform. In biology, NVIDIA and the Baker Lab at the University of Washington’s Institute for Protein Design said BioNeMo RFD4-Proteina generates protein sequences and 3D structures and delivers leading performance across more than 10 protein design tasks, with designs about 13 times faster than RFdiffusion3 in a benchmark of 200 amino acid proteins, according to NVIDIA. For quantum computing, NVIDIA said the Ising family is intended to support calibration and error correction across qubit modalities; it said IQM Quantum Computers is using Ising to calibrate processors, SEEQC is combining NVQLink and Ising, and AWS is integrating Ising into AWS Parallel Computing Service.
The company also listed availability for several models. NVIDIA said Nemotron 3.5 Super VL is available on Hugging Face, ModelScope, OpenRouter and build.nvidia.com as an NVIDIA NIM microservice, along with NVIDIA cloud, post-training and inference partners. It said Nemotron 3.5 Bolt E4B 4GB is expected on Friday, Oct. 23, and that Nemotron Retriever, Kumo, Cosmos, BioNeMo RFD4-Proteina and Ising are available through a mix of Hugging Face, GitHub, build.nvidia.com and partner platforms. NVIDIA said Alpamayo 2 Edge will be available later this month.
The blog post follows NVIDIA’s Oct. 15 and Oct. 14 Nemotron announcements, which added Super VL, Bolt and the Nemotron 3.5 Super and Lightning models. Those technical posts said Nemotron 3.5 Super has 120 billion total parameters and 12 billion active parameters, while Lightning has 30 billion total parameters and 3 billion active parameters; both support context lengths up to one million tokens. NVIDIA said the performance results in those materials are vendor benchmarks rather than independent verification.
As of Oct. 21, 2026 at 09:48 UTC, NVIDIA was priced at $216.47, down 0.35% on the day, with a market capitalization of $5,256.97 billion, according to the market data provided.