NVIDIA Nemotron 3 Nano Omni multimodal model now available on Amazon SageMaker JumpStart

NVIDIA announced day zero availability of Nemotron 3 Nano Omni, an open multimodal large language model with 30 billion total parameters and 3 billion active parameters, on Amazon SageMaker JumpStart. The model combines video, audio, image, and text understanding in a single architecture.

The model supports a 131K token context length and processes video (up to 2 minutes, up to 256 frames in mp4 format), audio (up to 1 hour, 8kHz+ sampling rate in wav or mp3), images (JPEG, PNG in RGB), and text in one inference pass. It is built on a Mamba2 Transformer Hybrid Mixture of Experts architecture and available in FP8 precision on SageMaker JumpStart.

Enterprise use cases include computer use agents that read screens and navigate graphical user interfaces, document intelligence for contracts and financial documents, and audio and video understanding agents for meeting analysis and customer service workflows. The approach eliminates the need for separate vision, speech, and language models stitched together.

Deployment is available through SageMaker JumpStart via one-click deployment in SageMaker Studio or programmatically using the SageMaker Python SDK. The model is licensed under the NVIDIA Open Model Agreement for commercial use.