Skip to content

Megrez-3B-Omni

Understand image, text and audio on device

Visit official site

What Megrez-3B-Omni is

The model is described as an on-device omni-modal understanding model built by extending Megrez-3B-Instruct to handle image, text and audio data.

What you can do with it

  • Understand and analyse images, text and audio together
  • Ask spoken questions about an uploaded image
  • Respond with text from a voice instruction
  • Hold multi-turn dialogue in Chinese and English speech

Getting started

  • Install the needed environment such as torch and transformers
  • Download the model from Hugging Face
  • Set the model path and load it with the code example
  • Prepare text, image and audio inputs

Pros & cons

✓ What we like

  • Chinese and English speech with multi-turn dialogue
  • Designed with response speed and edge deployment in mind

! What to watch out for

  • Spoken questions about images need a prepared image input
  • The listed benchmark scores come from named test sets

FAQ

Which modalities are covered?

The material lists image, text and audio understanding.

What speech is supported?

The material states Chinese and English speech input and multi-turn dialogue.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review