Nano Banana input tokens fall: what changes in an editing budget?

What changed

On November 4, 2025, Google reduced Gemini 2.5 Flash Image input-image token use from 1,290 to 258 tokens per image. That is an 80% reduction in this input component, not an 80% reduction in the total image bill.

Break the request into components

An editing budget should distinguish source images, text instructions, generated output, and any other applicable charges. A change to one component affects workflows differently. A reference-heavy editing task can have a different cost profile from generation using only a short prompt.

For an illustrative request with four source images, this change takes the image-input count from 5,160 to 1,032 tokens. It says nothing by itself about output tokens or your negotiated rate.

Update estimates carefully

Keep the original estimate and record which assumption changed. Recalculate at a matching workload and output setting. Avoid comparing one period with many large outputs against another period with fewer small images and attributing the entire difference to token accounting.

After a configuration change, compare estimates with actual request logs. If they differ, inspect units and included components before assuming the provider rate is wrong.

Explain the saving accurately

For internal budget notes, use a sentence such as: image-input token consumption decreased for this model; overall savings depend on the request mix. That wording helps finance understand the change without promising a blanket discount.