In the rapidly evolving landscape of artificial intelligence, enterprises are turning to agentic AI to help supercharge productivity, automate complex workflows, and drive intelligent decision-making. At Oracle, we remain committed to providing our customers access to leading reasoning models through both OCI Generative AI and OCI Data Science. Today, we’re thrilled to announce that OCI Data Science is making another family of enterprise-grade models available, furthering our commitment to customer choice. The NVIDIA Nemotron family of models can be enabled on OCI Data Science Model Deployment, providing a new option for open-source multimodal reasoning models. These models are designed to support a wide range of enterprise agentic tasks with leading compute efficiency. Furthermore, NVIDIA has used open-source data to train the models to provide full transparency and offers tools to easily customize and deploy the models, empowering businesses to adopt them with greater visibility and control.
Greater Value for Enterprises, More Options for Access
Model deployments are a managed resource in the OCI Data Science service used to deploy models as HTTP endpoints in OCI. Customers can clone the NVIDIA NIM container from NVIDIA NGC to Oracle Cloud Container Registry (OCIR) and then deploy the Nemotron models to the OCI Data Science Model Deploy service. This allows customers to use these models in their own workflows, leveraging Oracle’s robust cloud infrastructure for scalable, secure AI development. After the models are deployed, OCI Data Science handles all infrastructure operations, including compute provisioning and load balancing for the model deployment.
Step-by-step instructions on how to access and deploy these models in OCI Data Science.
The Rise of Agentic AI and the Need for Advanced Reasoning
Enterprises are increasingly adopting agentic AI to help streamline operations and boost efficiency. As AI agents evolve from simple task executors to autonomous systems, reasoning models will become the cornerstone of their intelligence. These models enable agents to perceive, understand, and act across multiple modalities, including text, images, audio, video, and structured data.
However, adopting existing reasoning models can present significant challenges. Many top models are proprietary, limiting customization and control over data. Open-source models, while offering flexibility, can suffer from inconsistent quality and be subject to security vulnerabilities, which erode trust. Building custom models from scratch demands enormous expertise, high-quality data, and massive compute resources, creating high barriers for most organizations.
The NVIDIA Nemotron family of models address these pain points and deliver enterprise-grade reasoning and greater trust at a lower barrier to entry for organizations. Built on popular open-source foundation models like Llama, Mistral, and Qwen, Nemotron models are then post-trained with NVIDIA’s high-quality synthetic data and advanced techniques, including reinforcement learning for human-like reasoning. This results in a family of models that, according to NVIDIA, think fast and explore deeper and more diverse reasoning paths to achieve better outcomes.
Nemotron models are optimized for various enterprise needs:
- Nano: Offering the highest efficiency and lowest latency, perfect for edge agents that need to deliver real-time automation.
- Super: Delivers the highest accuracy and leading throughput on a single NVIDIA H100 GPU—ideal for balanced performance in mid-scale deployments.
- Ultra: Increased accuracy for complex, multi-agent workflows, such as customer service automation, supply chain management, and IT security in data centers.
Build, train, deploy, and manage machine learning (ML) models
OCI Data Science provides customers with a streamlined, efficient environment to work with foundational models from a variety of model providers. Users can choose from a growing catalog of enterprise-ready models or import any LLM from OCI Object Storage, then fine-tune and deploy via an easy-to-use interface.