Skip to content

Bunny

Lightweight multimodal models with pluggable encoders

Visit official site
Category AI models

What Bunny is

The project is described as a family of lightweight multimodal models with plug-and-play vision encoders and language backbones.

What you can do with it

  • Combine image and text understanding
  • Add image understanding to a chatbot
  • Use the models as a multimodal backend

Model notes

  • EVA-CLIP and SigLIP are named as encoders
  • Llama-3-8B and Phi-1.5 are named as backbones
  • Bunny-v1.0-3B is compared with 7B and 13B models

Pros & cons

✓ What we like

  • Several encoder and backbone choices are offered
  • Models can be found on HuggingFace and ModelScope

! What to watch out for

  • Richer training data is used to offset smaller size
  • The named models centre on a 3B and an 8B variant

FAQ

Which encoders are named?

EVA-CLIP and SigLIP are listed as vision encoder options.

Where are model details found?

HuggingFace, ModelScope and other distribution platforms are named.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review