NVIDIA Introduces a Family of NVIDLA Llama Nemotron Language Open Reasoning AI Models to Help Businesses and Developers Create Agentic AI Platforms
In order to give developers and businesses a business-ready platform for developing sophisticated AI agents that can operate alone or in coordinated teams to accomplish challenging tasks, NVIDIA today unveiled the open Llama Nemotron family of models with reasoning capabilities.
The NVIDIA Llama Nemotron reasoning family provides on-demand AI reasoning capabilities and is based on Llama models. In order to increase multistep math, coding, reasoning, and complicated decision-making, NVIDIA improved the new reasoning model family during post-training.
This refinement technique optimises inference time by 5x when compared to other top open reasoning models and increases model accuracy by up to 20% when compared to the basic model. The models can now handle increasingly difficult reasoning tasks, increase decision-making skills, and lower operating expenses for businesses to the gains in inference performance.
NVIDIA’s new reasoning models and software are being developed in partnership with leading agent AI platform pioneers, such as Accenture, Amdocs, Atlassian, Box, Cadence, Crowd Strike, Deloitte, IQVIA, Microsoft, SAP, and Service Now.
“The adoption of reasoning and agentic AI is amazing,” stated NVIDIA founder and CEO Jensen Huang. “Developers and businesses worldwide can build an accelerated agentic AI workforce with NVIDIA’s open reasoning models, software, and tools.”
NVIDIA Post-Training Improves Enterprise Reasoning Accuracy and Reliability
The Llama Nemotron Language model family, which is designed to provide production-ready AI reasoning, is offered as NVIDIA NIM micro services in Nano, Super, and Ultra sizes, each of which is tailored to meet specific deployment requirements.
The Ultra model will provide maximum agentic precision on multi-GPU servers, the Super model offers the best accuracy and throughput on a single GPU, and the Nano model provides the highest accuracy on PCs and edge devices.
Using high-quality curated synthetic data produced by NVIDIA Nemotron and other open models, together with additional curated datasets co-created by NVIDIA, NVIDIA carried out significant post-training on NVIDIA DGX Cloud.
Businesses will have the freedom to create their own unique reasoning models since the tools, datasets, and post-training optimisation strategies used to create the models will be publicly accessible.
NVIDIA and Agentic Platforms Collaborate to Improve Industry Reasoning
Leaders in the agentic AI platform market are collaborating with the Llama Nemotron thinking models to provide businesses with sophisticated reasoning.
Microsoft is incorporating NIM microservices and Llama Nemotron Language reasoning models into Microsoft Azure AI Foundry. This adds more possibilities for customers to improve services like Azure AI Agent Service for Microsoft 365 in the Azure AI Foundry model catalogue.
To enhance SAP Business AI products and Joule, the AI copilot from SAP, the company is utilising Llama Nemotron Language models. Furthermore, it is promoting improved code completion accuracy for SAP ABAP programming language models by utilising NVIDIA NIM and NVIDIA NeMo microservices.
To increase company efficiency across industries, Service Now is using Llama Nemotron models to create AI agents with improved performance and accuracy.
Accenture’s AI Refinery platform now offers NVIDIA Llama Nemotron Language reasoning models, along with new industry agent solutions that were unveiled today. This allows clients can quickly create and implement custom AI agents that are suited to industry-specific problems, speeding up business transformation.
The recently launched Zora AI agentic AI platform from Deloitte, which aims to enable and mimic human decision-making and action with agents that have deep functional and industry-specific business knowledge and built-in transparency, will include Llama Nemotron reasoning models.
NVIDIA AI Enterprise Provides Crucial Agentic AI Tools
To accelerate the adoption of sophisticated reasoning in collaborative AI systems, developers can use the new NVIDIA agentic AI tools and software to use NVIDIA Llama Nemotron Language thinking models.
The most recent agentic AI building blocks are all included in the NVIDIA AI Enterprise software platform and include:
- Businesses can connect knowledge to AI agents that can see, think, and act on their own to the NVIDIA AI-Q Blueprint. The blueprint, which was constructed with NVIDIA NIM microservices, incorporates NVIDIA NeMo Retriever for multimodal information retrieval and uses the open-source NVIDIA AgentIQ toolkit to provide agent and data connectivity, optimisation, and transparency.
- The AI-Q Blueprint was used to create the NVIDIA AI Data Platform, a configurable reference design for a new class of enterprise infrastructure incorporating AI query agents.
- New NVIDIA NIM microservices allow for real-time adaption in any environment and continuous learning, while optimising inference for complex agentic AI applications. The most recent models from top model builders, such as Microsoft, Mistral AI, and Meta, are reliably deployed to the microservices.
- AI agents may continuously learn from feedback provided by both humans and AI to NVIDIA NeMo microservices, which offer an effective, enterprise-grade solution for rapidly establishing and maintaining a robust data flywheel. With the help of NVIDIA microservices, developers will be able to create and optimise data flywheels with ease to the reference architecture provided by the NVIDIA AI Blueprint.
Accessibility
Hugging Face and build.nvidia.com offer the NVIDIA Llama Nemotron Nano and Super models with NIM microservices as a hosted application programming interface. Members of the NVIDIA Developer Program have free access to development, testing, and research resources.
With NVIDIA AI Enterprise, businesses may use cloud and enhanced data centre infrastructure to operate Llama Nemotron NIM microservices in production. When NVIDIA NeMo microservices become available to the general public, developers can register to get notifications.
It is anticipated that the NVIDIA AI-Q Blueprint will be accessible in April. You may now access the NVIDIA AgentIQ toolbox on GitHub.
