The price, resolution by resolution
Google released Nano Banana 2.1 on 6 October, an image generation and editing model built on Gemini 3.6 Flash. Its published pricing bills output images at $30 per million tokens, against $60 for the gemini-3.1-flash-image model it replaces.
That works out differently at each size. A 1K image costs $0.0336, down from $0.067. A 2K image costs $0.0504, down from $0.101. A 4K image costs $0.113, down from $0.151 — a cut of about a quarter, not the half that applies lower down the range.
The old model is being switched off on 29 October, so this is a migration with a deadline rather than an optional upgrade.
What the model card claims
The model card puts the input context at 1 million tokens, with image output capped at 4K tokens and text output at 64K. Google says the model handles up to 14 reference images and holds consistency across four characters and ten objects.
On Google’s own comparison against Gemini 3.1 Flash, overall preference scores 1050 ± 14 against 990 ± 7, multi-character consistency 1106 ± 14 against 978 ± 10, and infographic factuality 0.521 against 0.179. These are vendor figures on a vendor comparison; no independent evaluation has been published.
The limitations Google lists itself
The model card is unusually direct about what still fails: poor text rendering at small sizes, character rendering that is not always consistent, and occasional confusion around spatial localisation. Knowledge for some domains cuts off in March 2026.
That last point matters more for an image model than it sounds. Infographic factuality is one of the three numbers Google leads with, and an infographic is only as factual as the knowledge behind it.
Where it runs
Distribution is the part Google does not have to argue for. The model is going out through the Gemini app, Google AI Studio, the Gemini API, Google Search’s AI Mode and Google’s enterprise platform, alongside Flow, Stitch and Google Ads.
That is a wide default surface, and it is the reason a 50% price cut on a Flash-tier image model is worth more attention than the benchmark table. Most of the images this model generates will be generated by people who never chose it.
What to watch
The 29 October shutdown is the date to hold. Anything still calling gemini-3.1-flash-image after that will break, and the replacement is cheaper at 1K and 2K but noticeably less cheap at 4K — which is the resolution anyone doing print or large-format work is using. Whether that pricing shape holds, or whether Google brings 4K down to match, is the thing to watch next.
