NVIDIA Nemotron 3 Super Launches on Amazon Bedrock

A New Era for Agentic AI: NVIDIA Nemotron 3 Super Arrives on Amazon Bedrock

Hey there, AI enthusiasts! Big news just dropped: NVIDIA's powerful new Nemotron 3 Super model is now fully available and managed on Amazon Bedrock. If you've been looking to supercharge your multi-agent applications and build more sophisticated AI systems, this is definitely something to get excited about. It’s designed from the ground up to offer top-tier compute efficiency and accuracy, especially for complex, multi-agent workflows. No more wrestling with infrastructure; Amazon Bedrock handles all that for you, letting you focus on building amazing generative AI applications.

This isn't just another model; Nemotron 3 Super is specifically engineered to excel in scenarios where multiple AI agents need to collaborate or where highly specialized reasoning is required. Think of it as bringing a whole new level of intelligence and efficiency to your AI projects, all while being easily accessible and managed.

What is Nemotron 3 Super? A Quick Rundown

So, what exactly makes Nemotron 3 Super so special? At its heart, it's a hybrid Mixture of Experts (MoE) model featuring a unique Hybrid Transformer-Mamba architecture. This fancy combination means it's built for serious computational power and precision. It’s particularly adept at multi-agent applications, enabling more sophisticated and specialized AI systems than ever before. Developers will be thrilled to hear that it supports a generous token budget, which translates directly to improved accuracy, especially when dealing with nuanced reasoning tasks.

One of the coolest aspects for developers is that Nemotron 3 Super is released with open weights, datasets, and recipes. This means you have the flexibility to customize, improve, and deploy the model on your own infrastructure, giving you enhanced privacy and security options. It’s all about empowering you to build exactly what you need with cutting-edge tools.

Deep Dive: Why Nemotron 3 Super Stands Out

Let's get into the nitty-gritty of why Nemotron 3 Super is a game-changer. For starters, it boasts the highest throughput efficiency in its size category, delivering up to a 5x improvement over the previous Nemotron Super model. This means faster processing and more bang for your buck. When it comes to performance, it offers leading accuracy for reasoning and agentic tasks among open models, with up to 2x higher accuracy compared to its predecessor. This isn't just marketing talk; it achieves high scores across a range of leading benchmarks, including AIME 2025, Terminal-Bench, SWE Bench (both verified and multilingual), and RULER.

Under the hood, Nemotron 3 Super packs a punch with a size of 120 billion parameters, while efficiently using 12 billion active parameters. It also supports an impressive context length of up to 256K tokens, allowing it to process and understand incredibly long and complex inputs. What's more, it’s a multilingual powerhouse, supporting English, French, German, Italian, Japanese, Spanish, and Chinese, making it suitable for a global audience.

Key architectural innovations include Latent MoE, which lets the model call on 4x more experts at the same inference cost. This results in better specialization for subtle semantic structures, domain abstractions, and multi-hop reasoning. Then there’s Multi-token prediction (MTP), a feature that significantly increases throughput for long reasoning sequences and structured outputs by predicting several future tokens in a single pass. For tasks like planning, trajectory generation, extended chain-of-thought, or code generation, MTP drastically reduces latency and boosts agent responsiveness. If you're keen to explore more technical details and how it's trained, you can Run Nemotron 3 Super on Amazon Bedrock to see it in action.

Real-World Applications

The potential for Nemotron 3 Super spans across numerous industries and use cases. Here are just a few examples of how developers and teams can leverage this powerful model:

Software Development: Imagine speeding up tasks like code summarization, making development workflows smoother and more efficient.
Finance: Accelerate loan processing by accurately extracting data, analyzing income patterns, and even detecting fraudulent operations, all leading to reduced cycle times and lower risk.
Cybersecurity: Use it to triage security issues, perform in-depth malware analysis, and proactively hunt for threats, bolstering your organization's defenses.
Search: Enhance search capabilities by deeply understanding user intent, allowing for more relevant results and the activation of the right agents.
Retail: Optimize inventory management, provide real-time personalized product recommendations, and improve in-store service, creating a better customer experience.
Multi-Agent Workflows: This is where Nemotron 3 Super truly shines. It can orchestrate task-specific agents—handling planning, tool use, verification, and domain execution—to automate complex, end-to-end business processes.

The ability to tackle such a diverse array of tasks with improved efficiency and accuracy makes Nemotron 3 Super a versatile tool for driving innovation.

Getting Started with Nemotron 3 Super on Amazon Bedrock

Ready to dive in? Accessing Nemotron 3 Super is straightforward, thanks to its integration as a fully managed, serverless model on Amazon Bedrock. This means you don't need to worry about provisioning or managing underlying infrastructure; you just use the model.

You can easily test the model using the Amazon Bedrock console's Chat/Text playground. Just navigate there, select NVIDIA from the model category list, and choose NVIDIA Nemotron 3 Super. After applying, you can start experimenting!

For those who prefer programmatic access, you can interact with the model using the AWS CLI or AWS SDKs (like Boto3 for Python) via the nvidia.nemotron-super-3-120b model ID. This allows you to integrate Nemotron 3 Super seamlessly into your existing applications and workflows. For detailed instructions on how to get started, including code examples for both the console and programmatic access, you'll find everything you need in the Amazon Bedrock announcement.

Read more: Run Nemotron 3 Super on Amazon Bedrock to begin building with this powerful new AI model today.