Nvidia just unveiled its new Llama Nemotron family of open models with reasoning capabilities, providing developers and enterprises with a business-ready foundation for creating advanced AI agents. The Nvidia Llama Nemotron family is built on Llama models, but Nvidia enhanced this new reasoning model family during post-training to improve multistep math, coding, reasoning, and complex decision-making.

Nvidia has been a frontrunner in the AI space for a long time, but with the arrival of DeepSeek R1, completion has been high. While the green team sees DeepSeek as a fantastic and world-class open-source reasoning model, they also believe that their Llama Nemotron family can do it even better – not necessarily in terms of being the smartest but in making their AI products more easily adaptable for businesses.

3 different sizes for 3 different purposes

The Llama Nemotron model family is available as NVIDIA NIM microservices in three different sizes: Nano, Super, and Ultra, each optimized for different deployment needs. The Nano model delivers the highest accuracy on PCs and edge devices, the Super model offers the best accuracy and highest throughput on a single GPU, and the Ultra model provides maximum agentic accuracy on multi-GPU servers.

Thanks to the refinement of Llama models, the Nvidia Llama Nemotron family delivers up to 20% better accuracy compared to the base model and optimizes inference speed by 5x compared to other leading open reasoning models. This means the models can handle more complex reasoning tasks, enhance decision-making capabilities, and reduce operational costs for enterprises.

“Better” than DeepSeek R1, says CEO Jensen Huang

During the keynote, Jensen Huang, CEO of Nvidia, said that while DeepSeek’s R1 model is much smarter, they believe they “can do it even better” by making these models more easily integrated into enterprise workflows. Additionally, these models are optimized for Nvidia hardware and can run efficiently on DGX Spark, DGX Station, and other Nvidia-powered cloud and server setups. This means that, unlike DeepSeek’s R1, Nvidia is making the case that its model is inherently designed with enterprise applications in mind, and is natively supported on their AI infrastructure.

“We are collaborating with NVIDIA to integrate Llama Nemotron reasoning models into Joule to enhance our AI agents, making them more intuitive, accurate and cost effective.”

Walter Sun, Global head of AI at SAP, source: Nvidia

Leading AI platform pioneers, including Accenture, Box, CrowdStrike, Deloitte, and SAP, have already collaborated with Nvidia on its new reasoning models and software to enhance their AI agents. On top of that, similar to DeepSeek’s R1, these models will be open source, meaning their tools, datasets, and post-training optimization techniques will be openly available, giving enterprises the flexibility to build their own custom reasoning models.

About the Author

Hassam Nasir

Hassam boasts over seven years of professional experience as a dedicated PC hardware reviewer and writer.