Unlocking the Future of AI with Nemotron 3 Super on Amazon Bedrock

Introduction

Explore the capabilities of the fully managed, serverless Nemotron 3 Super model, designed to revolutionize generative AI applications with unrivaled efficiency and accuracy.

Unleashing Innovation with NVIDIA Nemotron 3 Super on Amazon Bedrock

The tech landscape is ever-evolving, and the recent launch of NVIDIA Nemotron 3 Super as a fully managed and serverless model on Amazon Bedrock is set to propel generative AI applications into new realms of efficiency and capability. This integration joins the already available Nemotron Nano models within the Amazon Bedrock ecosystem, making sophisticated AI solutions accessible to developers without the burden of managing infrastructure complexities.

What Makes Nemotron 3 Super Stand Out?

Architectural Brilliance

At the heart of Nemotron 3 Super lies a hybrid Mixture of Experts (MoE) architecture that employs cutting-edge Transformer-Mamba designs. This allows for:

• Token budget management that enhances accuracy while minimizing reasoning tokens.

Unmatched Performance

With a size of 120 billion parameters, the model boasts incredible throughput efficiency, achieving:

• 5x the throughput efficiency of its predecessor, the Nemotron Super.
• Up to 2x higher accuracy for reasoning and agentic tasks compared to earlier versions.

Moreover, extensive benchmarks like AIME 2025 and Terminal-Bench validate its capability across multiple languages, including English, French, German, Italian, Japanese, Spanish, and Chinese.

Innovative Features

1. Latent MoE: This approach allows the model to utilize four times more experts without increasing inference costs, resulting in a finely tuned specialist around complex semantics and multi-hop reasoning patterns.

2. Multi-token Prediction (MTP): MTP enables the model to predict several future tokens in one go, significantly enhancing throughput for extended reasoning sequences and structured outputs.

For a deeper dive into its workings, check out the detailed insights in "Introducing Nemotron 3 Super: an Open Hybrid Mamba Transformer MoE for Agentic Reasoning."

Diverse Use Cases for Nemotron 3 Super

The capabilities of Nemotron 3 Super extend across various sectors, enabling innovation that drives real-world impact:

• Software Development: Automate code summarization and other development tasks efficiently.
• Finance: Expedite loan processing through data extraction and analysis, aiding in fraud detection.
• Cybersecurity: Enhance threat detection and perform detailed malware analyses.
• Search Optimization: Improve user intent understanding, triggering the right responses to queries.
• Retail Management: Optimize inventory and provide personalized recommendations in real-time.
• Multi-Agent Workflows: Automate complex business processes by orchestrating dedicated agents for specific tasks.

Getting Started with Nemotron 3 Super

Ready to test the remarkable capabilities of Nemotron 3 Super? Follow these simple steps:

1. Navigate to the Amazon Bedrock console.
2. Select Chat/Text playground from the left menu under the Test section.
3. Choose Select model in the upper left corner.
4. Pick the NVIDIA category and select NVIDIA Nemotron 3 Super.
5. Click Apply to load the model.

Testing the Model

To showcase the prowess of Nemotron 3 Super, challenge it with a complex engineering prompt. For instance:

"Design a distributed rate-limiting service in Python that must support 100,000 requests per second across multiple geographic regions."

This requires the model to engage in high-level system design, code implementation while addressing threading, race conditions, and including test cases.

Advanced Integration with AWS CLI and SDKs

Programmatic access to Nemotron 3 Super is straightforward:

Using the AWS CLI

Run the following command to invoke the model directly from your terminal:

aws bedrock-runtime invoke-model \
 --model-id nvidia.nemotron-super-3-120b \
 --region us-west-2 \
 --body '{"messages": [{"role": "user", "content": "Your Prompt Here"}], "max_tokens": 512, "temperature": 0.5, "top_p": 0.9}' \
 --cli-binary-format raw-in-base64-out \
invoke-model-output.txt

Using AWS SDK for Python (Boto3)

Here’s a quick script to interact with the model:

import boto3
from botocore.exceptions import ClientError

client = boto3.client("bedrock-runtime", region_name="us-west-2")
model_id = "nvidia.nemotron-super-3-120b"

user_message = "Your Prompt Here"
conversation = [{"role": "user", "content": user_message}]

try:
    response = client.converse(
        modelId=model_id,
        messages=conversation,
        inferenceConfig={"maxTokens": 512, "temperature": 0.5, "topP": 0.9},
    )
    print(response["output"]["message"]["content"][0]["text"])

except (ClientError, Exception) as e:
    print(f"ERROR: Can't invoke '{model_id}'. Reason: {e}")

Conclusion: Your Gateway to Advanced AI

This post highlights the capabilities of NVIDIA Nemotron 3 Super on Amazon Bedrock, revolutionizing agentic AI applications. With its sophisticated architecture and serverless framework, organizations can leverage high-reasoning applications without the complexities of backend management.

Ready to unleash the power of Nemotron 3 Super for your workflows? Explore the Amazon Bedrock Console and dive into the future of generative AI today!

About the Authors

Aris Tsakpinis
A Senior Specialist Solutions Architect for Generative AI, Aris combines professional expertise with ongoing PhD research in Machine Learning Engineering.

Abdullahi Olaoye
A Senior AI Solutions Architect at NVIDIA, Abdullahi specializes in integrating NVIDIA AI frameworks with cloud services to enhance AI model deployment and workflows.