Skip to content

Magma-8B

A multimodal foundation model for agent research

Visit official site

What Magma-8B is

The model is described as a multimodal foundation model built to study multimodal AI agents.

What you can do with it

  • Generate text conditioned on image or video
  • Produce a visual trajectory for a task
  • Ground a UI action such as a button click

Getting started

  • Install transformers, torch and other packages
  • Load the model and its processor
  • Prepare an image and a text prompt

Pros & cons

✓ What we like

  • Visual planning produces a task trajectory
  • UI grounding and robot control are listed

! What to watch out for

  • Text and image are the named input types
  • Learning uses unlabeled video data

FAQ

Who developed it?

The card states the model was developed by Microsoft.

What can it ground?

The card lists UI actions such as clicking a button and robot arm control.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review