Google launches Nano Banana 2.1 ahead of the older image API's earliest shutdown
The new image model supports 4K output, Google Search grounding and multi-image editing. Google's deprecation schedule lists October 29th as the earliest shutdown date for Gemini 3.1 Flash Image, its recommended replacement.
By Ryan Merket · Published
Primary source: Google AI Studio
Why it matters
Google is offering the image model through a developer API and a range of its consumer and business products, while giving API developers an approaching migration date. Teams using Gemini 3.1 Flash Image will need to check whether Nano Banana 2.1 works for their workloads before Google's listed earliest shutdown date.
Google introduced Nano Banana 2.1 on October 6th as a new image-generation and conversational-editing model. Google's developer documentation, last updated that day, identifies it as an update to Nano Banana 2, the product name for Gemini 3.1 Flash Image. Developers can try it in Google AI Studio.
The Gemini API documentation says the model accepts text, images, video and PDFs, and returns images and text. Developers can request 1K, 2K or 4K output, use Google Web and Image Search grounding, and configure the model's Thinking level.
Google's API documentation says multi-image fusion supports up to 14 reference images, with consistency for as many as four characters and fidelity for up to 10 objects. The documentation also describes improvements in visual quality, prompt adherence, text rendering and consistency across turns. Those are Google's descriptions of its model, not independent test results.
A model built to travel across products
Google's Nano Banana 2.1 model card lists the Gemini app, Google AI Studio, the Gemini API, Google Search AI Mode, Google Ads, Google Flow and Google Stitch as distribution channels. The API model ID is gemini-nano-banana-2.1.
Google has not published a standalone revenue or user figure for Nano Banana 2.1. The release materials therefore do not establish how widely the model is being used.
The model card shows the limits of the claim
Google's model card reports stronger results than Nano Banana 2 and Nano Banana Pro on the listed image-generation and editing evaluations. For text-to-image overall preference, Google reports a score of 1,050 +/- 14 with Thinking enabled, compared with 990 +/- 7 for Nano Banana 2 and 935 +/- 8 for Nano Banana Pro. The card also reports higher scores for multi-character consistency and multi-reference editing.
These are Google's evaluation results, not a settled ranking of image models. The model card describes internal evaluations and human preference tests; its scores do not establish how often the model will produce usable images for a particular task. Google also notes limitations in spatial localization, advanced world knowledge, 3D reasoning and factuality. It says some domains may reflect knowledge only through January 2025, despite a March 2026 stated cutoff for Gemini 3.6 Flash.
Google's materials give different accounts of the model's underlying version. The API documentation calls Nano Banana 2.1 an update to Gemini 3.1 Flash Image, while the model card says it is based on Gemini 3.6 Flash. The published pages do not resolve the discrepancy.
The launch also gives API developers a migration date to track. Google's deprecation schedule lists Gemini 3.1 Flash Image for shutdown on October 29th and names Nano Banana 2.1 as its recommended replacement. The schedule says listed shutdown dates are the earliest possible dates; Google says customers will be notified of the exact service end date in advance. Developers using the older model should test the replacement against their own workloads before that earliest date.