TL;DR
Google released Nano Banana 2.1, its image-generation and editing model, on October 6 — already rolling out in the Gemini app and Search's AI Mode. It renders up to 4K, holds a subject's look across repeated edits, and API prices are roughly half of Nano Banana 2. Developers take note: the old model shuts down October 29.
What Google shipped
On October 6, Google made Nano Banana 2.1 generally available — the successor to Nano Banana 2, the model behind the image features that pushed Gemini to the top of both app stores last year. Google says it "outperforms our previous models across the board, with notable leaps in visual design, mask-based editing, and subject consistency to help you create more natural-looking images."
Under the hood it is built on Gemini 3.6 Flash, and it shows: the same Flash-tier speed and cost profile, but with outputs up to 4K resolution, aspect ratios as wide as 8:1, and fixes for the tiling artifacts that plagued wide panoramas on earlier versions.
It's already in your pocket
You don't need an API key to try it. Nano Banana 2.1 is rolling out in the Gemini app and in Google Search's AI Mode — Google's Robby Stein says to "open the Google app & tap the banana icon under the Search box to get started creating." It is also live in Google Ads, AI Studio, Flow, and Stitch for builders.
The consistency upgrades are the practical win. The model takes up to 14 reference images in one session and tracks up to four characters and ten objects, so a person or product keeps looking like itself across a run of edits instead of drifting between prompts. It can also check Google Search before drawing real-world things — an event, a place, a product — so the details have a better shot at being right.
What the numbers say — and what they don't
Google's own preference tests put Nano Banana 2.1 at 1050 ELO for text-to-image, ahead of Nano Banana 2 at 990 and Nano Banana Pro at 935, with infographic accuracy jumping from 0.179 to 0.521. Those numbers come from Google's own testing — independent results were not available at launch, so treat them as a claim, not a verdict.
Early hands-on testing suggests the ceiling hasn't moved: reviewers found Nano Banana Pro still produces the more realistic, natural-looking images. Google itself frames 2.1 as "the more efficient counterpart" to Pro — the fast, cheap workhorse, not the flagship.
The price cut and the deadline
For developers, the headline is cost. A 1K image drops from $0.067 to $0.0336; a 4K image from $0.151 to $0.0756 — roughly half, with batch jobs discounted another 50%. At 4K for under eight cents, high-resolution generation is suddenly cheap enough to use casually.
But move soon if you build on Google's image API. The old gemini-3.1-flash-image model — the one still called Nano Banana 2 — is deprecated and shuts down on October 29, 2026. Any integration pointing at it stops working then, so the switch to gemini-nano-banana-2.1 is a real deadline, not a suggestion.