Google halved the price of each AI image, and tripled what it charges to read your inputs
On 6th Oct'26, Google released Nano Banana 2.1, the image model built on Gemini 3.6 Flash, in the Gemini API and the Gemini app. A standard 1K image now costs about $0.034 through the API, half the previous model's price, but input tokens, the text and reference images you send in, rise from $0.50 to $1.50 per million, and text output rises from $3 to $7.50. There is no free tier on the API.
What it actually means
The saving is real if your workflow is a short prompt that comes back as a finished image, which is how most teams generating ad variants or product shots use it. The bill moves the other way when you send a lot in, because the model now accepts up to 14 reference images per prompt and each one is charged at three times the old rate. A brand team that feeds in a product shot and a dozen style references to get one image back may find its cost per finished asset barely moves, or rises.
Before you switch, take a week of your real requests and price them both ways, counting the tokens going in as well as the images coming out. If most of your jobs can wait, Google's Batch tier halves both sides again, and that is the bigger saving for anyone producing images at volume.
The price cut is on what comes out, and the price rise is on what goes in.
3xGoogle's new input price for its image model, against the old one
Source: Google Gemini API pricing · Thursday 8 October 2026