Summary
- On October 6, Google introduced Nano Banana 2.1, making it available through the Gemini application, AI Mode in Google Search, Google Ads, and various developer tools.
- Internal assessments suggest it surpasses both Nano Banana 2 and Nano Banana Pro, achieving an infographic factuality score of 0.521 compared to 0.179, although independent evaluation was not conducted at the time of release.
- The pricing for developers is approximately 50% less than that of Nano Banana 2, starting at $0.0336 for a standard 1K image.
Google formally launched Nano Banana 2.1 on Tuesday, the latest iteration of its AI model designed to transform text into images or modify existing photos.
This version is being integrated into the Gemini app, the AI Mode of Google Search, Google Ads, and various developer tools such as Google AI Studio, Flow, and Stitch.
The original Nano Banana propelled Gemini to success, and by September 2025, its feature that converted selfies into collectible figurines helped Gemini achieve the top spot on major app stores, surpassing ChatGPT after nearly three years.
During this period, Alphabet's market capitalization also exceeded $3 trillion.
The second version was launched in February, based on Gemini 3.1 Flash Image. Its significant improvement involved verifying information through Google Search before generating images, enhancing the accuracy of depictions of real people or events.
Introducing Nano Banana 2.1, our latest model for image generation and editing š
This upgraded version excels compared to our previous models, showing significant improvements in visual design, mask-based editing, and consistency of subjects to create more realistic images. pic.twitter.com/d858DrzrF3
ā Google (@Google) October 6, 2026
What enhancements come with this update? Google highlights three key areas: improved visual design, enhanced mask-based editing (allowing users to modify specific areas), and better subject consistency, which ensures that a person or object retains its identity through multiple edits.
In text-to-image preference tests, Nano Banana 2.1 achieved 1,050 ELO Points, outperforming Nano Banana 2 at 990 and Nano Banana Pro at 935. The ELO system operates without a ceiling, making it a qualitative measure of performance.
This model supports up to 14 reference images simultaneously, tracking four characters and ten objects. It can produce images with a resolution up to 4K, providing TV-level clarity, and supports aspect ratios as wide as 8:1.
Developers have the flexibility to adjust the model's "thinking" time prior to generating an image, from minimal to extensive, and can enable it to reference Google Search and Google Image Search for accuracy before creating visuals.
Via Google's developer API, which allows external applications to utilize the model, the cost for a standard 1K image is $0.0336, roughly half of the $0.067 charged for Nano Banana 2.
A 4K image is priced at $0.0756 compared to $0.151 for the previous version. Additionally, batch processing, which allows for the bulk handling of images, offers a further 50% discount.
At the standard 1K rate, processing a thousand images now costs about $33.60 through the API, a significant reduction from the $67 for Nano Banana 2, making this model not only highly efficient but also cost-effective.