# NVIDIA Releases Efficient Nemotron-3 Multimodal MoE

> The new 30-billion parameter Mixture-of-Experts model handles text and images while using only 3 billion active parameters for inference.

Published by The Open Weights on Apr 24, 2026. Canonical: https://theopenweights.com/news/nemotron-3-nano-omni-30b-a3b-reasoning-cbno

## Key facts

- Company: NVIDIA
- Model: Nemotron-3 Nano Omni 30B-A3B Reasoning
- Version: 3
- Category: Any-to-Any
- Modalities: Any-to-Any, Reasoning, Vision-Language
- License: Other (Open weights, commercial use allowed)
- Parameters: 30B (mixture of experts)
- Active parameters: 3B
- Context window: 1.0M tokens
- Architecture: NemotronH_Nano_Omni_Reasoning_V3
- Precision: INT8
- Weights size: 20.6 GB
- Languages: English
- Significance: notable
- Published: 2026-04-24
- Last verified: 2026-06-17
- Hugging Face: https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4
- Canonical URL: https://theopenweights.com/news/nemotron-3-nano-omni-30b-a3b-reasoning-cbno

NVIDIA has released Nemotron-3 Nano Omni, a new multimodal model designed for complex reasoning tasks. The model continues the company's expansion into open AI, offering a sophisticated tool capable of understanding and processing both text and visual information.

The key innovation in Nemotron-3 Nano Omni is its efficient architecture. It's a Mixture-of-Experts (MoE) model with a total of 30 billion parameters, but it only activates a fraction of them—just 3 billion—for any given task. This design significantly reduces the computational cost of inference compared to a traditional dense model of the same size, a focus highlighted by the "Nano" in its name.

As a Vision-Language Model (VLM), Nemotron-3 Nano Omni can perform tasks that require a simultaneous understanding of images and language. This makes it suitable for applications like detailed image captioning, visual question answering, and other reasoning challenges that depend on integrating visual context with textual prompts.

Developers can access the model, optimized in the NVFP4 format, on the [Hugging Face Hub](https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4). It is available under the Nvidia Open Model License, a custom license that users should review before integrating it into their projects. This release provides a powerful, resource-conscious option for teams building multimodal AI applications.

## Get the model

- [Hugging Face](https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4)

## Sources

- [nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4](https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4) — Hugging Face, Apr 24, 2026

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "NVIDIA Releases Efficient Nemotron-3 Multimodal MoE", The Open Weights, Apr 24, 2026, https://theopenweights.com/news/nemotron-3-nano-omni-30b-a3b-reasoning-cbno