SunsettingAI Image

Nano Banana 2 (gemini-3.1-flash-image)

by Google

Announced

2026-10-06

Complete

2026-10-29

Refund window closes

—

Refund status

No refund applicable

On October 6, 2026, Google's Gemini API release notes announced Gemini Nano Banana 2.1 (`gemini-nano-banana-2.1`) as generally available and, in the same entry, deprecated Nano Banana 2 (`gemini-3.1-flash-image`) with a shutdown date of October 29, 2026. That is 23 days of notice for a model that only went GA on May 28, 2026. It is also the second Nano Banana migration in a week for some teams: the deprecations page lists the original Nano Banana, `gemini-2.5-flash-image`, with a shutdown date of October 2, 2026, and the pricing page warning for that model told users to migrate to Gemini 3.1 Flash Image, the model now being retired. The replacement is described by Google as an update to Nano Banana 2 and is documented in the same image generation guide. The pricing is not a straight swap: generated images cost about half as much, but input and text output cost more, and the 512px size is not supported.

Refund flow

1
Not applicable. Nano Banana 2 in the Gemini API is billed per token on the paid tier, with no pre-paid balance tied to a model ID, so there is nothing to refund when the model is retired.
2
If you see charges for gemini-3.1-flash-image calls dated after the model stops serving, treat it as a billing error and raise it through Google Cloud billing support with the request details.
3
Re-budget before you switch, using the prices on Google's pricing page on October 6, 2026. Image output on gemini-3.1-flash-image is $60 per 1M tokens, listed as $0.045 per 512px image, $0.067 per 1K, $0.101 per 2K and $0.151 per 4K. Image output on gemini-nano-banana-2.1 is $30 per 1M tokens, listed as $0.0336 per 1K, $0.0504 per 2K and $0.0756 per 4K, so each generated image costs about 50% less at the same size. Input moves the other way, from $0.50 to $1.50 per 1M tokens, and text and thinking output from $3 to $7.50 per 1M tokens. Batch image output drops from $30 to $15 per 1M tokens. If your calls carry many reference images or long prompts, check your own token mix before assuming the switch saves money.

Migration path

Recommended successor

Gemini Nano Banana 2.1 (gemini-nano-banana-2.1)

Why this is the closest fit

It is the replacement Google names in the October 6, 2026 deprecation notice and on the deprecations page, and Google describes it as an update to Nano Banana 2 that keeps Flash-level speed. The image generation guide documents both models together and lists the same 1K, 2K and 4K token counts per image for each. Video-to-image generation, which the May 28, 2026 release notes described as exclusive to gemini-3.1-flash-image, is listed in the current guide as available on Nano Banana 2.1 too, as is mixing up to 14 reference images.

What differs from the original

Three things change besides the model name. First, size: the image generation guide says the 512px (0.5K) resolution is supported on Gemini 3.1 Flash Image and not on Nano Banana 2.1, so code that requests image_size "0.5K" for thumbnails or drafts needs a new size. Second, price shape: image output is about half the cost per image, while input and text output cost 3x and 2.5x as much per token. Third, aspect ratios: the release notes list wide and panoramic 1:4, 4:1, 1:8 and 8:1 ratios at 1K, 2K and 4K among the 2.1 improvements. Google also claims better visual quality, prompt adherence, character consistency across turns and text rendering. Those are Google's claims and were not tested for this record.

Alternatives by use case

You generate 1K to 4K images with short prompts

Switch to gemini-nano-banana-2.1 before October 29, 2026. At listed prices each output image costs about half. Run your own prompts on both before the cutoff

You use 512px output for thumbnails or cheap drafts

Nano Banana 2.1 has no 512px size. Its 1K image is listed at $0.0336, below the $0.045 you paid for 512px on Nano Banana 2, so request 1K and downscale if you need small files

You edit with many reference images or long prompts

Input tokens triple from $0.50 to $1.50 per 1M. Log the input and output token counts from a few real calls on both models and compare the totals before you move a whole pipeline

You only need 1K images at the lowest cost

Google also lists gemini-3.1-flash-lite-image (Nano Banana 2 Lite), which only supports 1K and has no shutdown date announced. Check its current price and output on your own prompts

What this tool meant

Nano Banana 2 lasted five months as a GA model, from May 28 to October 29, 2026, and Google gave 23 days of notice. Teams that moved off gemini-2.5-flash-image on Google's own advice now have to move again. A GA label in the Gemini API is not a long-term commitment, so keep your prompts, reference images and a saved output for every asset type you care about, and re-run them on the replacement before the cutoff instead of trusting the release notes. If those images feed an AI video pipeline as first frames or references, check the new outputs inside the video step too, not only as stills.