GoogleGemma 3
Gemma 3 4B
Gemma 3's smaller vision-language checkpoint.
What the release provides
The 4B model accepts images and text, generates text, and has a shared 128K window. Images are encoded to 256 tokens.Google Gemma 3 model card
How to use it well
Practical guidance. Inspect small text, crowded scenes and document images separately. Keep image preprocessing fixed while comparing responses. Reserve enough context for the answer, and validate any extracted values against the original image before using them in a business operation.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | google/gemma-3-4b-it |