Google launches Nano Banana 2.1: sharper image edits at half the price — and a three-week shutdown notice
Google on Tuesday released Nano Banana 2.1, its latest image generation and editing model — roughly half the price of Nano Banana 2, ahead on every benchmark Google published, and already rolling out to the Gemini app and AI Mode in Search. Developers on the old model have until October 29 to switch.
Google announced Nano Banana 2.1 on Tuesday, its latest image generation and editing model — one it says "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." The model is built on Gemini 3.6 Flash, per Google's model card, and it replaces Nano Banana 2, which launched in February and reached general availability in May.
The rollout is broad: the Gemini app, AI Mode in Google Search, Google AI Studio, Flow, Stitch, Google Ads, and the Gemini Enterprise Platform. Robby Stein, Google's VP of Product for Search, said 2.1 "is rolling out in AI Mode in Search" — users tap the 🍌 icon under the search box in the Google app to start generating images directly.
Half the price
For developers, the headline is price. A standard 1K image drops to 3.36 cents from 6.70 cents under Nano Banana 2; a 4K image falls from 15.10 to 7.56 cents. Nano Banana Pro, the premium tier, stays put at 13.4 cents per 1K image. The model runs at Flash-tier speed, outputs up to 4K, supports extreme aspect ratios from 8:1 to 1:8 for panoramas, and can juggle up to 14 reference images while keeping as many as four characters and ten objects consistent across edits.
The benchmark sheet
Google's published numbers put 2.1 ahead of its predecessor — and in several categories, ahead of Pro itself. On text-to-image overall preference, 2.1 scored 1050 ELO points against 990 for Nano Banana 2 and 935 for Pro. Infographic accuracy jumped to 0.521 from 0.179, a notable leap on a task models are usually terrible at. Mask-based editing scored 1049 versus 965, and multi-character consistency hit 1106 against 978. There is independent corroboration, too: the arena.ai text-to-image board puts Nano Banana 2.1 fifth at 1328 ELO, about 80 points ahead of Pro at 1248 — though GPT Image 2.5 still holds the top two slots.
It leaked first
The launch caps a week of quiet telegraphing. On October 5, the tracking account TestingCatalog posted screenshots of "Nano Banana 2.1" sitting in Google Flow's model picker; the listing vanished within a day. The name had already been dialed back once — Flow's build previously pointed to "Nano Banana 2.5 Flash" before it was relabeled 2.1, a choice that telegraphed an incremental release rather than a generational jump. Google confirmed that reading: 2.1 is "the more efficient counterpart" to Pro, not its replacement.
Why it matters
The original Nano Banana turned image generation into a cultural event last year, when its collectible-figurine trick pushed the Gemini app to the top of both app stores; version 2 added real-world grounding via Google Search. With 2.1, Google is doing the thing incumbents do when they sense price competition: ship a better, cheaper default and retire the old one. Developers running gemini-3.1-flash-image have a hard deadline — the API turns it off on October 29.
One caveat from early hands-on testing: multi-turn editing degrades. After five or six successive edits in a single thread, faces soften and small details drift — the practical workflow is to restart from a clean generation rather than stack a long edit chain. And in practice, reviewers still find Pro produces noticeably better final images when quality is worth paying for. Nano Banana 2.1 is the workhorse.