Newsletter

Google's own tests show that Nano Banana 2.1 is ahead of Nano Banana 2 and Nano Banana Pro. One of these infographics has a factual accuracy score of 0.521, while the latter two have scores of 0.179 respectively; however, there are no independent test results available at the time of publication.

Developers are priced at about half of Nano Banana, with a starting price of $0.0336 for standard 1K images.

Google released Nano Banana 2.1 on Tuesday, which is the latest version of this AI model that can convert input text into images or edit your existing photos.

The initial Nano Banana made Gemini popular. In September 2025, it added the feature of turning selfies into collectible figurines, which propelled Gemini to the top of both major app stores, ending ChatGPT's leading position that had lasted for nearly three years.

During the same period, the market value of Alphabet exceeded $3 trillion.

Version 2.0 was launched in February and is built upon Gemini 3.1 Flash Image. One of its main features is that it checks Google Search before drawing real-world objects, which means that pictures of real events or people are more likely to be drawn accurately.

So what are the new features in this update? Google has listed three improvements: a better visual design, more precise mask-based editing functionality (you mark an area, and only that area will be modified), and stronger consistency of subjects, meaning that characters or objects maintain their original characteristics after multiple edits.

In terms of overall preference for text-to-image conversion tests, Nano Banana 2.1 received 1,050 ELO points. Nano Banana 2 got 990 points, and Nano Banana Pro scored 935 points. In the ELO system, the higher the score, the better. There is no upper limit to the scores, so it is merely a qualitative indicator.

From the paper parameters, it can handle up to 14 reference images simultaneously, track up to 4 characters, and 10 objects at a time. It can output at a maximum resolution of 4K, which is roughly equivalent to television-grade clarity, with a maximum aspect ratio of 8:1.

Developers can also set the duration for the model to "think" before drawing, ranging from the lowest to the highest, and allow it to use Google Search and Google Image Search to grounding the results. In layman's terms, it can look up information before creating the drawing.

Through Google's developer API, which is the channel that allows other companies' applications to call this model, the price for a standard 1K image is 0.0336 US dollars. This is approximately half of the 0.067 US dollars for Nano Banana 2.

The price of a 4K image is $0.0756, while Nano Banana 2 costs $0.151. For bulk tasks, which involve processing a large number of images at once, you can enjoy an additional 50% discount.

Calculated at the standard price of 1K, generating 1000 images through API now costs approximately $33.60, while the cost with Nano Banana 2 is $67. This makes this model "extremely outstanding," as well as very efficient and cost-effective.