Open Models Almanac
Methodsmodel architecturemultimodal architecture

implementation detail · filed under model architecture

Configurable visual token budget

Gemma 4 lets the caller cap how many tokens an image becomes, at 70, 140, 280, 560 or 1,120, and resizes the image to fit. A 1024×1024 image becomes 64 to 1,089 tokens across the five settings (computed from the processor code); higher budgets keep more detail and cost more context.

Written for the explainer Getting images and video into a decoder: encoders, projectors, and native fusion.

sources
3
model
1
lab adopt it
1
strongest
optional

Documented in

Google DeepMind

Evidence

3 quotes from 3 sources, one card per model, each opening with its strongest.

How sources treat it
optional 3

Gemma 4 31B

Google DeepMindadopted3 sources

Gemma 4 supports variable image resolution through a configurable visual token budget, which controls how many tokens are used to represent an image.

optionalinference servingin Gemma 4Google DeepMind
2 more from this model

Gemma 4 supports variable image resolution through a configurable visual token budget, which controls how many tokens are used to represent an image.

optionalinference servingin Gemma 4Google DeepMind

Gemma 4 supports variable image resolution through a configurable visual token budget, which controls how many tokens are used to represent an image.

optionalinference servingin Gemma 4Google DeepMind
↑