Skip to content

moondream

A small vision-language model built on SigLIP and Phi-1.5

Visit official site
Category AI models

What moondream is

The model is described as a 1.6 billion parameter vision-language model built from SigLIP, Phi-1.5 and the LLaVA training dataset.

What you can do with it

  • Try the model on Huggingface Spaces
  • Review the weights under the CC-BY-SA license
  • Read the numbers listed for VQAv2, GQA, VizWiz and TextVQA

Model notes

  • Weights are covered by CC-BY-SA
  • The material prints benchmark numbers for several models
  • The LLaVA dataset is named as the reason for the license

Pros & cons

✓ What we like

  • The model can be tried on Huggingface Spaces
  • Numbers are listed for VQAv2, GQA, VizWiz and TextVQA

! What to watch out for

  • Weights are bound by the CC-BY-SA license
  • The listed scores are the numbers printed in the material

FAQ

How large is the model?

The material states 1.6 billion parameters.

Why does CC-BY-SA apply?

The material says the weights are covered because the LLaVA dataset is used.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review