Nvidia unveiled a new suite of artificial intelligence (AI) models on Tuesday during its GPU Technology Conference (GTC) 2025. Named Llama Nemotron, these models are tailored to emphasize reasoning capabilities and are aimed at providing a foundational platform for agentic AI workflows. The Santa Clara-based company elaborated that these models are designed for developers and enterprises seeking to create advanced AI agents that can function autonomously or collaborate in teams to tackle complex objectives. The Llama Nemotron models can now be accessed through Nvidia’s platform and Hugging Face.
Nvidia Unveils New Reasoning-Focused AI Models
In a recent announcement, Nvidia provided insights into the newly launched AI models. The Llama Nemotron reasoning models are built upon Meta’s Llama 3 series and include enhancements integrated by Nvidia following their initial training. The company emphasized that these models exhibit superior performance in areas such as multistep mathematics, coding, reasoning, and making complex decisions.
Nvidia revealed that enhancements during the training process improved model accuracy by up to 20 percent compared to their predecessor models. Additionally, the speed of inference has been accelerated fivefold when compared to other open-source reasoning models of similar size. According to Nvidia, “the models can tackle more intricate reasoning challenges, improve decision-making, and help businesses lower operational costs.” These improvements position the Llama Nemotron models as a valuable tool for developing and operating AI agents.
The Llama Nemotron reasoning models come in three distinct parameter sizes: Nano, Super, and Ultra. The Nano model is optimized for on-device and edge applications where high accuracy is important. The Super model provides a balance of high accuracy and throughput suitable for single GPU usage, while the Ultra variant is designed for deployment on multi-GPU servers, offering top-tier accuracy.
Post-training of these reasoning models occurred on the Nvidia DGX Cloud using specially curated synthetic data generated through the Nemotron platform and additional open models. Nvidia also announced their intention to share the tools, datasets, and optimization techniques used in developing the Llama Nemotron models with the open-source community.
Furthermore, Nvidia is collaborating with enterprise partners to facilitate access to these models for developers and businesses. The reasoning models, along with NIM microservices, are available through Microsoft’s Azure AI Foundry, as well as via Azure AI Agent Services. SAP is incorporating these models in its Business AI solutions and the AI copilot known as Joule. Other firms utilizing the Llama Nemotron models include ServiceNow, Accenture, and Deloitte.
The Llama Nemotron Nano and Super models, along with NIM microservices, can be accessed by developers and businesses as an application programming interface (API) through Nvidia’s platform and its Hugging Face listing. They are available under the Nvidia Open Model License Agreement, which permits both research and commercial applications.
