NVIDIA Nemotron 3 Super: A Leap Forward in Agentic AI

NVIDIA has launched Nemotron 3 Super, a 120-billion-parameter open-weight model designed to power complex agentic AI systems at scale. The model, now available, combines advanced reasoning capabilities with improved efficiency for autonomous agents, addressing key challenges in multi-agent AI workflows.

Addressing Context Explosion and the ‘Thinking Tax’

As companies move beyond simple chatbots to more sophisticated multi-agent applications, two primary constraints emerge. The first is “context explosion,” where multi-agent workflows generate significantly more tokens – up to 15 times more than standard chat – due to the require to resend full interaction histories, including tool outputs and reasoning steps. This increased context volume drives up costs and can lead to “goal drift,” where agents lose focus on the original objective. The second challenge is the “thinking tax,” the computational expense of complex agents reasoning at every step.

Nemotron 3 Super: Key Features and Architecture

Nemotron 3 Super tackles these challenges with a 1-million-token context window, enabling agents to retain full workflow state and prevent goal drift. The model achieves up to 5x higher throughput and 2x higher accuracy compared to the previous Nemotron Super model through a hybrid architecture incorporating three key innovations:

• Mamba Layers: Deliver 4x higher memory and compute efficiency.
• Transformer Layers: Drive advanced reasoning capabilities.
• Hybrid Mixture-of-Experts (MoE): Activating only 12 billion of its 120 billion parameters during inference.

Further enhancing performance is Latent MoE, a new technique that improves accuracy by activating four expert specialists for the cost of one token generation. Multi-Token Prediction, which predicts multiple future words simultaneously, results in 3x faster inference.

Performance Benchmarks and Optimization

Nemotron 3 Super has demonstrated leading performance, achieving the top spot on Artificial Analysis for efficiency and openness among models of the same size. It also powers the NVIDIA AI-Q research agent to the No. 1 position on both the DeepResearch Bench and DeepResearch Bench II leaderboards, benchmarks measuring an AI system’s ability to conduct thorough, multi-step research across large document sets while maintaining reasoning coherence.

Running on the NVIDIA Blackwell platform in NVFP4 precision, the model reduces memory requirements and achieves up to 4x faster inference than FP8 on NVIDIA Hopper, without sacrificing accuracy.

Open Weights, Data, and Accessibility

NVIDIA is releasing Nemotron 3 Super with open weights under a permissive license, allowing developers to deploy and customize it on workstations, in data centers, or in the cloud. The model was trained on synthetic data generated using frontier reasoning models, and NVIDIA is publishing the complete methodology, including over 10 trillion tokens of pre- and post-training datasets, 15 training environments for reinforcement learning, and evaluation recipes. Researchers can further fine-tune the model or build their own using the NVIDIA NeMo platform.

Applications in Agentic Systems

Nemotron 3 Super is designed for complex subtasks within multi-agent systems. Examples include:

• Software Development: Loading an entire codebase into context for end-to-end code generation and debugging.
• Financial Analysis: Processing thousands of pages of reports to improve efficiency and eliminate the need for repeated reasoning.
• Cybersecurity: Ensuring high-accuracy tool calling for autonomous security orchestration.

Availability and Deployment Options

NVIDIA Nemotron 3 Super is accessible through several platforms:

• Direct Access: build.nvidia.com, Perplexity, OpenRouter, and Hugging Face.
• Dell Technologies: Available on the Dell Enterprise Hub on Hugging Face, optimized for on-premise deployment on the Dell AI Factory.
• HPE: Bringing NVIDIA Nemotron to its agents hub for scalable enterprise adoption.

Enterprises and developers can also deploy the model through:

• Cloud Service Providers: Google Cloud’s Vertex AI, Oracle Cloud Infrastructure, and coming soon to Amazon Web Services through Amazon Bedrock and Microsoft Azure.
• NVIDIA Cloud Partners: Coreweave, Crusoe, and Together AI.
• Inference Service Providers: The basset hound, CloudFlare, DeepInfra, Fireworks AI, Inference.net, Lightning AI, Modal, and FriendliAI.
• Data Platforms and Services: Distilled, Dataik, DataRobot, Deloitte, EY, and Tata Consultancy Services.

The model is packaged as an NVIDIA NIM microservice, enabling deployment from on-premises systems to the cloud.

Stay up to date on agentic AI and NVIDIA Nemotron by subscribing to NVIDIA AI news, joining the community, and following NVIDIA AI on LinkedIn, Instagram, X, and Facebook.