The industry’s push toward agent AI continues, with Nvidia announced several new services and models to facilitate the creation and deployment of AI agents.

Today, Nvidia launched Nemotron, a family of models based on Metaby Llama and trained in the company’s techniques and datasets. The company also announced new AI orchestration blueprints to guide AI agents. These latest releases bring Nvidia, a company best known for the hardware powering the generative AI revolution, to the forefront of agentic AI development.

Nemotoron comes in three sizes: Nano, Super and Ultra. It also comes in two flavors: the Llama Nemotoron for language tasks and the Cosmos Nemotoron vision model for physical AI projects. The Llama Nemotron Nano has 4B parameters, the Super 49B parameters and the Ultra 253B parameters.

All three are best for agent tasks including “following instructions, chat, function calling, coding and math,” according to the company.

Rev Lebaredian, VP of Omniverse and simulation technology at Nvidia, said in a briefing with reporters that the three sizes are optimized for different Nvidia computing resources. Nano is for cost-efficient low latency applications on PCs and edge devices, Super is for high accuracy and throughput on a GPU and Ultra is for maximum accuracy at data center scale.

“AI agents are the digital workforce that will work for us and work with us, and therefore the Nemotoron model family for agent AI,” Lebaredian said.

Nemotron models are available as API hosts on Hugging Face and Nvidia’s website. Nvidia says businesses can access the models through its AI Enterprise software platform.

Nvidia is no stranger to foundational models. Last year, it was quietly released a version of Nemotron, Llama-3.1-Nemotron-70B-Teachingwhich outperforms similar models from OpenAI and Anthropic. This too NVLM 1.0 is opena family of multimodal language models.

More support for agents

AI agents becomes a big trend in 2024 as businesses begin to explore how to deploy agent systems in their workflow. Many believe that the momentum will continue this year.

Companies want Salesforce, ServiceNow, AWS and Microsoft everyone is calling agents the next wave of gen AI in businesses. Added AWS multi-agent orchestration in Bedrock, while Salesforce released it Agentforce 2.0which brings many agents to its customers.

However, agent workflows still require other infrastructure to function efficiently. One such infrastructure revolves around orchestration, or managing multiple agents that cross different systems.

Orchestra blueprints

Nvidia has also entered the burgeoning field of AI orchestration with its blueprints that guide agents through specific tasks.

The company collaborates with many orchestral companies, including LangChain, LlamaIndex, CrewAI, Day and Weights and Biasesto create Nvidia AI Enterprise blueprints. Each orchestration framework creates its own blueprint with Nvidia. For example, CrewAI created a blueprint for code documentation to ensure code repositories are easy to navigate. LangChain adds Nvidia NIM microservices to its structured report generation blueprint to help agents return internet searches in a variety of formats.

“Making multiple agents work together seamlessly or orchestrate is the key to deploying agent AI,” Lebaredian said. “These leading AI orchestration companies integrate every Nvidia agentic building block, NIM, Nemo and Blueprints into their open-source agentic orchestration platform.”

Nvidia’s new PDF-to-podcast blueprint aims to compete The Google NotebookLM by converting information from PDF to audio. Another new blueprint helps build agents to search and summarize videos.

Lebaredian said Blueprints aims to help developers quickly deploy AI agents. To that end, Nvidia unveiled Nvidia Launchables, a platform that allows developers to test, prototype and run blueprints with one click.

The orchestra will be one of the bigger stories in 2025 as businesses grapple with multi-agent production.