As a part of the Nemotron family, Nvidia also announced Cosmos Nemotron vision language models.
The LLMs under the Nemotron family are available in Nano, Super, and Ultra sizes
Photo Credit: Nvidia
Nvidia announced the Llama Nemotron family of open large language models (LLMs) on Monday. The company said that with the rise of artificial intelligence (AI) agents, new and more sophisticated AI models were required to handle the workflow of agentic AI. Highlighting the need for more power and higher efficiency, the tech giant stated that the Nemotron family models can create and deploy AI agents across various applications. The company claimed that the AI models will be available for enterprises via the Nvidia NIM microservice.
In a blog post, the tech giant announced its new series of open-source LLMs dubbed Nemotron. The series also contains Cosmos Nemotron vision language models (VLMs), and these can be used to build AI agents that analyse and respond to images and videos. Nvidia said the vision-focused agents can be deployed in autonomous machines, hospitals, stores and warehouses, as well as sports events, movies, and news.
Built with Meta's Llama foundation models, the Nvidia Llama Nemotron models are said to be optimised to build and develop AI agents. While the company did not reveal the architecture and technical details, it claimed that these models are trained using “latest techniques and high-quality datasets”. The models can be used to train agentic capabilities such as instruction following, chat, function calling, coding and mathematics, and more. Nemotron is also said to optimise the AI agents' size to make it easy to deploy.
Nvidia stated that SAP, ServiceNow, and other AI agent platform providers will be among the first to use the new Llama Nemotron models.
The Nemotron and Cosmos Nemotron models will be available in three parameter sizes — Nano, Super, and Ultra. Nano is the most cost-effective model built with low latency as the primary focus. Super is a high-accuracy model that can be run on a single GPU. Finally, Ultra is the highest-accuracy model designed for data centre-scale applications.
Nvidia highlighted that enterprises can access the Nemotron model family as downloadable models and as NIM. These models will also be available as application programming interfaces (APIs). While the models are open-source, they are only available for academic and research usage.
Catch the latest from the Consumer Electronics Show on Gadgets 360, at our CES 2026 hub.