Skip to content

PaliGemma 2 mix

One vision-language model for many visual tasks

Visit official site

What PaliGemma 2 mix is

The model is described as a Google vision-language model in the Gemma family covering captioning, OCR, visual question answering, detection and segmentation. It offers pretrained checkpoints at 3B, 10B and 28B parameters that can be finetuned for vision-language tasks.

What you can do with it

  • Generate short and long captions and run OCR
  • Answer questions about an image and detect objects
  • Segment images and deploy in Vertex Model Garden

Application scenarios

  • Produce accurate captions for short videos
  • Retrieve key information from images by question answering
  • Support medical image analysis through segmentation

Pros & cons

✓ What we like

  • Upgrades from the original PaliGemma model without changes
  • Official documentation and sample code are provided

! What to watch out for

  • Checkpoint size and resolution must be picked before use
  • Model weights come from Kaggle or Hugging Face

FAQ

Which sizes are offered?

The material lists 3B, 10B and 28B checkpoints at 224px and 448px resolutions.

Which frameworks are named?

Hugging Face Transformers, Keras, PyTorch and JAX are listed as compatible.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review