Google Slashes Image Costs in Half With Nano Banana 2.1
Google's Nano Banana 2.1 halves per-image API prices and pushes Gemini image generation into Search, Ads and the Gemini app. Here's what changed and why.
Google has nearly halved what it charges for an API-generated image with Nano Banana 2.1, a Flash-tier model built on Gemini 3.6 Flash. Google says it outperforms Nano Banana 2 on all the measures the company reports. The model reached general availability on October 6, 2026, and it is already live inside the Gemini app, AI Mode in Google Search, Google AI Studio and the Gemini API. The headline figure is a price: a 1K image now costs $0.0336, against $0.067 on Nano Banana 2. Google is pricing image generation as a commodity, and it is asking the rest of the market to follow.
The cut is uniform across resolutions. Google's API pricing lists Nano Banana 2.1 at $0.0504 for a 2K image and $0.0756 for 4K output, both exactly half the rate charged by the outgoing model. On the standard paid Gemini API tier, image output is billed at $30 per million tokens instead of $60.
| Output size | Nano Banana 2.1 | Nano Banana 2 |
|---|---|---|
| 1K image | $0.0336 | $0.067 |
| 2K image | $0.0504 | $0.101 |
| 4K image | $0.0756 | $0.151 |
Google is distributing the model far beyond its developer platform. The rollout covers the Gemini app, AI Mode in Search, Google AI Studio, the Gemini API, Google Ads, Google Flow, Google Stitch and the Gemini Enterprise Agent Platform. That breadth is the part consumers will feel first: a half-price image model is now embedded in surfaces that never required an API key, from a search box to an ad console. On the Gemini Enterprise Agent Platform it sits alongside the higher-priced Pro tiers, so business buyers weigh the two on cost and fidelity.
Nano Banana 2.1 is cheaper, not Google's best
This release refreshes a Flash-tier line. It is not a new frontier system. Nano Banana Pro, which runs on the newer Pro model 3.1, remains Google's strongest image generator for everyday work. The complication is that 2.1 posts higher benchmark scores than Pro in some tests while undercutting Pro's 1K price by roughly three quarters. Benchmark leadership and real-world quality have come apart, and buyers now have to decide which signal to trust.
The capability set targets editing and consistency rather than raw fidelity. The model accepts up to 14 reference images in a single prompt, and Google states it can hold up to four characters and ten objects consistent across conversation turns. It supports conversational editing, three thinking levels (minimal, medium and high) and search grounding through Google Search and Google Image Search.
The editing tools gained a wider set of aspect ratios, which matters for banner and social formats where a fixed square or horizontal frame forces cropping. Google lists sharper mask editing and accurate text rendering among the upgrades, the two areas where earlier image models most often failed.
Search grounding is the consumer-facing feature that stands apart from the price story. By letting the model consult Google Search and Google Image Search before it renders, Google is targeting the failure mode that makes AI images unusable in commercial settings: details that look plausible but are factually wrong. For a product shot or a location render, grounding is what separates a draft from something a brand can publish.
The cadence is quick. Nano Banana 2 arrived in February 2026 under the API identifier gemini-3.1-flash-image, which puts this refresh about eight months later. Google is iterating on the cheap, fast image line faster than it refreshes the Pro tier, a signal of where it expects volume to sit. The price list makes the split explicit: the Flash line carries the volume, the Pro line carries the quality premium, and this release widened the gap between them.
The price cut has a second edge
The halving applies to image output, and only to image output. The cost of everything around the image moved the other way. Input tokens tripled from $0.50 to $1.50 per million, and text and thinking output rose from $3 to $7.50 per million, while the token count per image stayed flat. A workflow built on short prompts and single images still gets cheaper. One that feeds long prompts or heavy reasoning into the model will see the savings shrink, or reverse outright.
For high-volume users, the arithmetic is the point. A pipeline that produced 10,000 images a day on Nano Banana 2 would have spent about $670 on image output; the same run on Nano Banana 2.1 costs roughly $336. Google has made the image the cheap part of the request and left the surrounding tokens as the variable that decides the final bill.
Developers who want the true per-call cost need one more figure. Token consumption runs about 1,120 tokens per 1K input image, so the billed amount depends on the image size and the surrounding prompt together, not on resolution alone. The published per-image rates are the cleanest comparison Google offers, but they are not the whole invoice.
The three thinking levels give developers a second lever. Minimal, medium and high settings change how much reasoning the model spends before it renders, and thinking tokens carry the higher rate. A team chasing the lowest cost per image runs minimal; one that needs consistency across a long conversation pays for high.
Developers also face a migration clock. Google's shutdown notice for the outgoing model left a 23-day window to change a model string in production, which is a tight turnaround for teams running image pipelines at scale. The new API identifier is gemini-nano-banana-2.1.
Why this matters
Google has separated the price of an image from the price of the request that produces it. The image is now the cheapest line item in the pipeline, while the input and reasoning tokens around it carry the higher rates, so the $0.0336 figure reads as much as a marketing number as a cost. A rival that matches the sticker price on image output still faces the same token math, which means the per-image rate settles less of the total bill than it appears to. For buyers, the practical step is to model prompt length and thinking level alongside resolution; the open question is whether the Pro tier keeps its quality premium now that the cheaper line outscores it in some tests.
Related Articles
- Chrome Users in Asia-Pacific Gain Access to Advanced Gemini AI Tools
- Google Pics brings AI image creation and editing to Docs, Slides, and beyond
- Half-Price Gemini 3.7 Flash Turns Model Race Into Cost War
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.