Nano Banana 2.1 Price per Image: Half the Cost, Except at 4K

Nano Banana 2.1 Price per Image: Half the Cost, Except at 4K

On the Gemini API, Nano Banana 2.1 costs $0.0336 per 1K image, $0.0504 per 2K image and $0.113 per 4K image. At the official rate of 3.75 riyals to the dollar, that is about 13, 19 and 42 halalas. That is half the price of Nano Banana 2 at 1K and 2K, but at 4K the saving is only about 25%. Input now costs three times as much, so the more reference images you attach to a request, the smaller the gap gets.

Google released the model on October 6, 2026, generally available from day one, and marked Nano Banana 2 as deprecated, recommending a move to the new model. The headline that went around was half the price. The number is right, but not at every size, and not for every way of using it.

Nano Banana 2.1 price in riyals, next to Google's other image models

These are the per-image prices on the standard tier of the Gemini API, as published on Google's pricing page, with the cost of a thousand images in riyals beside each one:

Three things stand out in the table:

• At 1K and 2K the price really is half of Nano Banana 2. A 1K image on 2.1 also costs a quarter of a Pro image.
• A 1K image on 2.1 is cheaper than the smallest image on Nano Banana 2. Nano Banana 2 offered a 0.5K size at $0.045. Version 2.1 has no 0.5K size, but its 1K image already costs less than that.
• A 1K image on 2.1 costs the same as one on Nano Banana 2 Lite. Lite's output image is also $0.0336. The difference is on the input side: $0.25 per million tokens on Lite against $1.50 on 2.1.

With the Batch API, when you do not need the result right away, the price halves again: $0.0168 per 1K image, or SAR 63 per thousand images.

Why the 4K saving is not half

Google does not price images directly. It prices the tokens they consume. Output image tokens on 2.1 cost $30 per million, half of Nano Banana 2's $60. A 1K or 2K image consumes the same number of tokens on both models, 1,120 and 1,680, so the price comes out at exactly half.

A 4K image, however, consumes about 3,780 tokens on 2.1, against 2,520 on Nano Banana 2: half as many tokens again, at half the price per token. The result is $0.113 instead of $0.151, a saving of about 25%. If your plan relies on 4K images for print or billboards, budget on that figure, not on the half-price headline.

Reference images: input got more expensive

Editing is where 2.1 is strongest. You can attach up to 14 reference images, and it keeps up to 10 objects, such as your product, its packaging and your logo, looking like themselves across the images it generates. But the input price rose from $0.50 per million tokens on Nano Banana 2 to $1.50 on 2.1, and every image you attach counts as input.

Google's documentation puts an input image on Gemini 3 models at about 1,120 tokens at the default resolution, and fewer at lower resolution settings. On that upper estimate, each reference image costs about $0.0017 on 2.1, against about $0.0006 on Nano Banana 2. That is a small amount, but it adds up:

These figures leave out thinking tokens. The model has three thinking levels, with medium as the default, and thinking tokens are billed at $7.50 per million. Its best results in Google's tests came with thinking switched on. So the real number for your business comes from the usage report after your first hundred images, or from counting a request's tokens with count_tokens before you send it.

What this means for your store or business

If you produce images for an online store, for ads or for a restaurant menu, the question is not which model is smartest. It is which size and which method fits each job:

• Store product images: try 1K first. A product image on a store page or a phone screen often does not need 4K, and the gap between SAR 126 and about SAR 424 per thousand images is large on a catalog of thousands of items. View the images on your actual page, then decide.
• Designs with text: start at 2K. In the model card, Google acknowledges that small text often comes out blurry at 1K. If the image carries a price, a product name or a menu, generate it at 2K, or add the text afterwards in a design tool, and check every Arabic letter before publishing.
• Large catalogs: use batch. If you are generating backgrounds for a thousand products in one go, the Batch API brings a 1K image down to $0.0168.
• Wide banners. Version 2.1 fixes artifacts that used to appear in wide and tall formats (4:1, 8:1, 1:4 and 1:8) at 2K and 4K, the formats you need for the strip across the top of a store page.
• The image is a promise to the customer. If the model changes the color or shape of the packaging and nobody notices, you will pay for it in returns and complaints. Check every product image against a photo of the real product before publishing. To edit one part of a product photo without reshooting it, see our guide on editing product photos with AI without a reshoot.

Can Nano Banana 2.1 replace Nano Banana Pro?

In the human preference tests Google published in the model card, 2.1 with thinking scored higher than Nano Banana Pro on every listed category, including product consistency, multi-reference editing and infographic design. And its 1K image costs a quarter of a Pro image.

But these are Google's own tests of its own models, not an independent evaluation. The model card also lists limitations: small text, occasional left and right confusion, and character consistency between input and output images that is not always perfect. The decision should come from a trial on your own images, not from a results table.

If your system runs on Nano Banana 2

If a developer built you a tool that generates product or ad images through the Gemini API, first find out which model it calls:

• gemini-3.1-flash-image (Nano Banana 2): Google has deprecated it and recommends migrating to gemini-nano-banana-2.1. No shutdown date has been announced as of today. Moving now means half the price at the common sizes.
• If you use the 0.5K size: 2.1 does not offer it, so move to 1K, which already costs less.
• gemini-2.5-flash-image (the original Nano Banana): it shuts down on March 15, 2027, and Google's recommended replacement is Nano Banana 2 Lite, which we covered in our Nano Banana 2 Lite article.
• Test before you switch: run twenty images from your real work through both models, compare quality, response time and cost from the usage report, then switch.

The Origami view

We read Nano Banana 2.1 as the end of cost as an excuse for product imagery. At about 13 halalas an image, cost no longer decides how many images you try. The reviewer's time does. So the real investment now is in the review step: who matches the image against the product, who checks the Arabic text, and who approves publishing.

In the systems we build for stores and businesses, we do not hard-code image generation to a single model. The model name and the size are settings the system owner can change, and the cost of every image is recorded next to the product it was made for. When Google ships a new version, switching becomes a settings change and a short test, not a new project. This is part of what we offer in our services.

Key takeaways

• Nano Banana 2.1 price: $0.0336 per 1K image, $0.0504 per 2K image and $0.113 per 4K image, which is SAR 126, SAR 189 and about SAR 424 per thousand images.
• Half the price of Nano Banana 2 at 1K and 2K, but only about 25% cheaper at 4K.
• Input costs three times as much, so the saving shrinks as you add reference images.
• Generate designs that contain text at 2K, use batch for large catalogs, and check every product image before publishing.
• If your system runs on Nano Banana 2, plan the move now. No shutdown date has been announced yet, but it is no longer the recommended option.

Sources

• Google AI for Developers: Gemini API pricing: prices for Nano Banana 2.1, 2, Pro and 2 Lite, tokens per image size, and batch prices.
• Gemini API release notes (October 6, 2026): Nano Banana 2.1 general availability and the deprecation of gemini-3.1-flash-image.
• Gemini Nano Banana 2.1 model page: sizes, 14 reference images, thinking levels and batch support.
• Gemini API deprecations: the status of Nano Banana 2 and the shutdown date of gemini-2.5-flash-image.
• Gemini API media resolution: input image token counts for Gemini 3 models.
• Google DeepMind: Nano Banana 2.1 model card: distribution channels, preference test results and known limitations.
• Saudi Central Bank: SAMA affirms commitment to exchange rate policy: the official rate of 3.75 riyals to the dollar.