Dive Brief:
- Nvidia launched the Nemotron 3 family of open models, libraries and data geared toward powering transparent agentic AI development across industries, the company said on Monday.
- Nemotron 3 models come in three sizes, Nano, Super and Ultra. Nemotron 3 Nano is the most compute-cost-efficient member of the model family used for tasks such as software debugging and content summarization, according to the announcement.
- “With Nemotron, we’re transforming advanced AI into an open platform that gives developers the transparency and efficiency they need to build agentic systems at scale,” Nvidia Founder and CEO Jensen Huang said in the announcement.
Dive Insight:
Just as enterprises started to become used to the idea of agentic AI, a new breed of agents has come on the scene: multi-agent systems, which bring with them more complexity – and more promise – than their predecessors.
Multi-agent systems are emerging as a critical design pattern that enterprises will need to pay attention to, according to Gartner. Instead of a single agent tackling a discreet task independently, multi-agent systems are made up of many agents sometimes working together and sometimes competing against each other to achieve a shared goal. However, moving from single-model chatbots to collaborative multi-agent systems presents technical complexities for developers and introduces governance risks.
Indeed, while there are benefits to multi-agent systems such as greater flexibility and scalability, CIOs have to consider challenges including agent malfunction, complexity with coordination and unpredictable behavior, according to IBM.
Nvidia touted that Nemotron 3 addresses those challenges, pointing to early adopters such as Accenture, Synopsys, Zoom, Oracle and ServiceNow, and the work they’ve done to integrate Nemotron models to power AI workflows in manufacturing, software development, cybersecurity and other industries. The models’ hybrid mixture-of-experts architecture allows developers to build and deploy multi-agent systems at scale, the company said.
The Nemotron 3 models also support Nvidia’s sovereign AI efforts, the company said. The models allow companies in areas such as the European Union and South Korea to build AI systems aligned with their data and regional regulatory requirements.
The South Korean government has invested in building out its national AI infrastructure to accelerate enterprise and industry sovereign AI development, according to an October announcement with Nvidia. The country plans to deploy more than 50,000 Nvidia GPUs across its sovereign clouds and AI factories over the next several years, the announcement said.
