Nvidia’s AI assistant game is here with new models, orchestration plans

The industry’s push towards professional AI continues, and Nvidia announcing several new services and models to support the development and deployment of AI agents.

Today, Nvidia introduced Nemotron, a family of models based on Measure‘s Llama is trained in corporate technology and datasets. The company also announced new AI plans for the AI ​​assistant. This latest move brings Nvidia, a company best known for tools that enable the AI ​​revolution, to the forefront of AI development.

Nemotron comes in three sizes: Nano, Super and Ultra. It also comes in two versions: the Llama Nemotron for speech tasks and the Cosmos Nemotron vision model for physical AI projects. Llama Nemotron Nano has 4B parts, Super 49B parts and Ultra 253B parts.

All three work well for tasks that include “instructions, chats, phone calls, texting and math,” according to the company.

Rev Lebaredian, VP of Omniverse and simulation technology at Nvidia, said in a press briefing that the three sizes are optimized for different Nvidia computing products. Nano is a low-cost application for PC and peripheral devices, Super is for high accuracy and throughput on a single GPU and Ultra is for high accuracy at the data center level.

“AI assistants are digital workers who can work for us and work with us, so Nemotoron’s model family is for helpful AI,” Lebaredian said.

Nemotron models are available as host APIs on Hugging Face and Nvidia’s website. Nvidia said businesses can access these models through its AI Enterprise platform.

Nvidia is no stranger to basic models. Last year, it was quietly released type of Nemotron, Llama-3.1-Nemotron-70B-Recommendwhich won similar races from OpenAI and Anthropic. Also NVLM 1.0a family of multilingual models.

More support for sponsors

AI assistants it became a major trend in 2024 as businesses began to evaluate how to use their operating systems. Many believe that the power will continue this year.

Companies like Salesforce, ServiceNow, AWS and Microsoft both called agents the next gen AI in businesses. AWS has added multi-agent call at Bedrock, where Salesforce released it Agentforce 2.0bringing more agents to its customers.

However, workflows still require some tools to function properly. One of those things revolves around calling, or managing multiple agents that go through different processes.

Orchestration notes

Nvidia has also entered the emerging field of AI orchestration with its plans to guide agents through specific tasks.

The company has collaborated with several music companies, incl Results LangChain, LlamaIndex, Crew I, Every day and Weights and Biasesto develop plans for Nvidia AI Enterprise. Each music group has developed its own plans with Nvidia. For example, CrewAI created a codebase plan to ensure that storage is easy to navigate. LangChain has added Nvidia NIM microservices to its reporting architecture to help providers return web searches in a variety of ways.

“Making multiple agents work better together or on call is critical to deploying professional AI,” Lebaredian said. “These leading AI music companies are integrating every Nvidia platform, NIM, Nemo and Blueprints with their streaming media platforms.”

Nvidia’s new PDF-to-podcast app aims to compete Google NotebookLM for converting information from PDF to audio. Another new feature will help users to search and summarize videos.

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

Orchestration can be one of the art forms big news for 2025 as businesses struggle to create more agents.