Nano Banana 2.1 Pricing: Half the Cost at 1K, Not at 4K

Nano Banana 2.1 costs about half as much per image as Nano Banana 2 at 1K and 2K resolution, and about a quarter less at 4K. Google released the model on 6 October 2026, and its pricing page lists the figures below. The headline "50% cheaper" is true for the small sizes and not for the largest one, which is the size people pay most for.

This was written on launch week. Prices are in US dollars, taken from Google’s Gemini API pricing page on 7 October 2026, and they can change. It covers the API model. Prices inside the Gemini app are not listed there.

What Nano Banana 2.1 is

Nano Banana is Google’s name for its Gemini image generation and editing models. The API model for this release is `gemini-nano-banana-2.1`. Google’s image generation guide describes it as the high-efficiency workhorse for image generation and conversational editing, with improved visual quality, text rendering and consistency across turns. The changelog entry lists the same improvements, plus extended aspect ratios.

The Decoder, which covered the launch the same day, reports that the model is rolling out in the Gemini app, AI Mode in Search, AI Studio, Flow and Google Ads. Treat that list as press coverage. The developer documentation is the part to rely on for the API.

Price per image, side by side

These are the standard-tier per-image prices on the pricing page. The older model is `gemini-3.1-flash-image`, the one most people call Nano Banana 2.

• —1K image: $0.0336 for Nano Banana 2.1, against $0.067 for the older model. About half.
• —2K image: $0.0504 against $0.101. About half.
• —4K image: $0.113 against $0.151. About 25% lower, not half.

The batch tier, which trades speed for a lower price, follows the same shape. A 1K image is $0.0168 against $0.034, a 2K image is $0.0252 against $0.050, and a 4K image is $0.0567 against $0.076. The percentages are in the sums, so check them yourself: 0.113 divided by 0.151 is about 0.75.

One caution about secondary coverage. At least one write-up quotes a 4K price that does not match the pricing page. When two sources disagree on a number, use the vendor’s page and note the date you read it.

Where your bill may not halve

Per-image prices are only part of the bill. The same page lists input at $1.50 per million tokens for Nano Banana 2.1, against $0.50 for the older model. Image output tokens drop from $60 to $30 per million. The per-image figures already include the output side, but anything you send in, such as long prompts or reference images, is billed at the input rate.

So a request that sends several reference images may save less than half. How much less depends on how many tokens those images use, which this post has not measured. If you run image generation at volume, price one real request, not the table.

The pricing page also lists the free tier as not available for this model. If you were trying the older model for free in the API, that is a change worth knowing about before you migrate.

What the docs say you get

• —Output sizes of 1K, 2K and 4K. The 0.5K size is not available for this model.
• —Up to 10 images of objects and up to 5 images of characters as references, per the image generation guide.
• —Three thinking levels: minimal, medium (the default) and high.
• —Aspect ratios from 1:1 and 4:5 to 9:16, 16:9 and 21:9.
• —A SynthID watermark on every generated image.

The changelog also says `gemini-3.1-flash-image` is deprecated and tells developers to migrate. It gives no shutdown date. Some secondary sources quote a specific date for that shutdown. Google’s own entry does not, so plan for a deadline without assuming one.

Which size to use

If the picture ends up on a phone screen or a slide, 1K is usually enough, and that is where the saving is largest. 4K is for print and for crops you plan to zoom into. Before paying for 4K, check whether you need it. A short guide to AI image prompts is a cheaper place to start than regenerating at a higher size.

The cheaper model does not remove the usual problem: you still have to describe what you want clearly. Writing a prompt that works first try saves more generations than any price cut, and a photo prompt for ChatGPT shows the same habits on a different tool. A prompt enhancer can help if you start from a vague idea.

How this compares with other launches

The pattern is the one from Meta Muse pricing: the headline number is real, but the plan, the size and the tier decide what you pay. Whatever the model, check an AI answer when you are not the expert. That applies to a generated chart or infographic as much as to text, because legible text in an image is not the same as correct text.

The short version

Half price at 1K and 2K, a quarter off at 4K. Price a real request before you migrate.

Nano Banana 2.1 is cheaper per image, and the old model is deprecated without a stated end date. The size you generate at decides how much of the saving you see. Coursium teaches people to use AI at work, including how to brief an image tool and check what comes back. You can see how Coursium works if that is useful.