Run NVIDIA Nemotron 3 Super on Amazon Bedrock

This article announces the availability of NVIDIA Nemotron 3 Super on Amazon Bedrock, a fully managed, serverless model for building agentic AI applications.

• Nemotron 3 Super is a 120B parameter hybrid Mixture of Experts model with 12B active parameters
• Supports up to 256K token context length with multi-token prediction for improved throughput
• Achieves 5x higher throughput efficiency and 2x better reasoning accuracy than previous version
• Supports seven languages: English, French, German, Italian, Japanese, Spanish, and Chinese
• Use cases include software development, finance, cybersecurity, search, retail, and multi-agent workflows
• Accessible via Amazon Bedrock console, AWS CLI, AWS SDK, and OpenAI-compatible API
• Model ID: nvidia.nemotron-super-3-120b for programmatic access

Organizations can now deploy high-reasoning, efficient agentic AI applications at scale without managing infrastructure through Amazon Bedrock's fully managed service.