NVIDIA brings Nemotron 3 Nano Omni to Amazon SageMaker JumpStart on release day

On April 28, 2026, NVIDIA added the multimodal model Nemotron 3 Nano Omni to Amazon SageMaker JumpStart on release day. For teams on AWS, this shortens the path from a new model's announcement to a pilot: the service is already ready for deployment and inference.

What the model is

Nemotron 3 Nano Omni is an open multimodal LLM with 30 billion total parameters and 3 billion active ones. It is built on a hybrid Mamba2 Transformer Hybrid Mixture of Experts architecture. NVIDIA assembled the model from three parts: the Nemotron 3 Nano language core, the CRADIO v4-H vision encoder for images and video, and the Parakeet speech encoder for audio.

The model accepts video, sound, images and text as input and returns a text response. According to AWS's description, it is designed not only for chat but also for agentic scenarios. It supports a context of up to 131 thousand tokens, reasoning, tool calling, JSON responses and word-level timestamps for transcription.

In SageMaker JumpStart, the model is available in FP8, which favors a balance between quality and efficiency. The license also matters for commercial use: Nemotron 3 Nano Omni is distributed under the NVIDIA Open Model Agreement.

The model is meant to "see, hear and reason" across several modalities

in a single inference pass.

Where the model fits

The core idea of the announcement is to eliminate the zoo of separate models for vision, speech and text. In a typical enterprise agent system, each such module adds latency, complicates orchestration and breaks the shared context. AWS and NVIDIA propose using Nemotron 3 Nano Omni as a single perception layer: the model reads the screen, understands documents, transcribes speech and analyzes video, while the rest of the agent's logic runs on top of one consistent picture.

• Computer-use agents that navigate interfaces, dashboards and the browser
• Document intelligence for contracts, SOWs, financial documents, tables and screenshots
• Analysis of calls, meetings and other audio and video content in customer support
• Verification of visual events, such as deliveries or orders, where OCR and temporal context are needed

The model's input limits are fairly clear, and they already look practical for pilots. Video: MP4 up to 2 minutes and up to 256 frames; audio: WAV or MP3 up to one hour; images: JPEG and PNG; text: up to 131 thousand tokens. This is not a universal, limitless all-in-one tool, but for internal assistants, review pipelines and operational automation the range is more than workable.

Finally, AWS separately claims up to 9 times higher throughput compared with alternative open omni models.

How to launch the model

SageMaker JumpStart presents this release as a one-click deployment. The basic scenario is simple: open SageMaker Studio, go to the JumpStart section, find Nemotron 3 Nano Omni, select the model card and click Deploy. Beforehand, AWS asks you to check three things: that you have an account, access rights to JumpStart and quotas for GPU instances, such as ml.p4d.24xlarge or ml.p5.48xlarge. So a quick start exists, but it still depends on the enterprise's infrastructure readiness and GPU budget.

For teams that deploy models through code, there is also a path via the SageMaker Python SDK with a ready-made model_id. After deployment, the endpoint accepts multimodal requests: you can describe an image, summarize a recording of a meeting or transcribe a call and extract action items.

AWS also recommends two inference modes: thinking for complex reasoning with temperature 0.6, top_p 0.95 and max_tokens 20480, and instruct for more direct tasks where speed matters more. After experimenting, it is best to delete the endpoint right away to avoid accumulating unnecessary costs.

What this means

Nemotron 3 Nano Omni arriving in JumpStart on release day shows that AWS is speeding up the delivery of fresh open models straight into the production environment. For business, this is a good signal: multimodal agents are gradually moving from a set of scattered components to a more cohesive product stack that can be tested on your own data without a long build from scratch.