Google Halved Nano Banana Image Prices at 1K and 2K. Input Now Costs 3x. What Does an Ad Variant Cost Now?

Google halved the image meter. Then it raised the other two.

On Oct 6 2026 Google made Nano Banana 2.1 generally available in the Gemini API and deprecated Nano Banana 2 (Gemini API release notes). The pricing page cuts image output from $60 to $30 per million tokens. The same page raises input from $0.50 to $1.50 per million tokens and text and thinking output from $3 to $7.50 (Gemini API pricing).

If your team generates ad variants, event creative, or one-to-one account images through Google's models, the headline is half price. Your bill depends on how much text, reference material, and thinking rides along with each image.

The rate card

• Image output. Nano Banana 2.1: $0.0336 at 1K, $0.0504 at 2K, $0.113 at 4K. Nano Banana 2: $0.067, $0.101, $0.151 (Gemini API pricing).
• Input. $1.50 per million tokens for text, image, and video on 2.1. Nano Banana 2 lists $0.50. That is 3x.
• Text and thinking output. $7.50 per million tokens on 2.1. Nano Banana 2 lists $3. That is 2.5x.
• Batch. Half of every 2.1 rate: $0.0168 at 1K, $0.0252 at 2K, $0.0567 at 4K.
• Free tier. None for 2.1 on the Developer API.
• Grounding. 5,000 free search requests a month, shared across Gemini 3.x models, then $14 per 1,000. Google says one request may run more than one search.

The 4K price moved on launch day

Google bills an image by its token count. At 1K a 2.1 image is 1,120 tokens. At 2K it is 1,680. At 4K the pricing page now says 3,780 tokens, or $0.113 (Gemini API pricing).

That 4K number changed after launch. Kingy AI logged the correction at 9:25 p.m. PDT on Oct 6: the launch-day page listed 2,520 tokens for a 4K image, and the revised page lists 3,780 (Kingy AI). At $30 per million, 2,520 tokens is about $0.0756. That figure was still going around after launch (Scoy). The real 4K cut is 25%.

One more 4K wrinkle. Nano Banana 2 on Batch lists $0.076 at 4K. Nano Banana 2.1 on Standard lists $0.113. If you already run 4K on Batch, moving to 2.1 Standard raises your image line by about half. Move to 2.1 Batch at $0.0567 instead.

What an ad variant costs

Take one ad variant. 2,000 input tokens for the brief, brand rules, and copy. 1,000 thinking tokens. One image out. Reference images and earlier edit turns add input tokens on top, at the new $1.50 rate.

• 2K on Nano Banana 2. $0.001 input + $0.003 thinking + $0.101 image = $0.105. $1,050 per 10,000 variants.
• 2K on Nano Banana 2.1. $0.003 + $0.0075 + $0.0504 = $0.0609. $609 per 10,000. A 42% cut.
• 4K on Nano Banana 2. $0.001 + $0.003 + $0.151 = $0.155. $1,550 per 10,000.
• 4K on Nano Banana 2.1. $0.003 + $0.0075 + $0.113 = $0.1235. $1,235 per 10,000. A 20% cut.

Now let the model think harder. At 5,000 thinking tokens per variant the 2K cut drops to 22% and the 4K cut drops to 8%. Ignore input, and about 8,400 thinking tokens per 4K image wipes out the saving. At 2K it takes about 11,200.

Google's model spec lists three thinking levels, minimal, medium, and high, with medium as the default (Kingy AI; model page). Google has not published typical thinking token counts per image. Read them off your own responses before you set a budget.

Same lesson as our GPT-6 and Opus 5.5 token cut decoder: the unit price fell, and the bill depends on the units you burn. For another meter with fine print under a cheap headline, see Claude Haiku 5.5's 100K token cliff.

Where this hits GTM

• Paid social and display variants. High volume, usually 1K or 2K. The cut is real here. Put it on Batch and set thinking to minimal.
• One-to-one account creative. Each image carries account context and reference images. Input at 3x eats into the saving. Measure input tokens per request.
• Event and print assets at 4K. The smallest cut, and the line that moved on launch day. Price on your own requests.
• Creative tools you already pay for. If a tool you pay for runs Google image models behind its own credits, a cut in Google's rate does not change your contract. Ask.

Nano Banana 2 is on the clock, with no date

Google deprecated gemini-3.1-flash-image on Oct 6 and says no shutdown date is announced (Gemini API release notes). Kingy AI reports an Oct 29 shutdown date was listed and then removed (Kingy AI). Plan the move now. Ask for a notice period before you depend on either model.

Google also offers project-level and billing-account spend caps in AI Studio (Gemini API release notes). Set one on the creative project before a bulk run. Our OpenAI usage tiers decoder covers why the cap needs a named owner.

Order Form lines for Nano Banana 2.1

1. Price per image in writing. By resolution: 1K, 2K, 4K, Standard and Batch. The token rate alone moved the 4K price on launch day.
2. Token count per resolution. Lock 1,120, 1,680, and 3,780 tokens for the term, or get notice before any change.
3. Input and thinking rates. $1.50 input and $7.50 text and thinking per million tokens. Budget the whole request.
4. Thinking level. Set minimal for bulk variants. Test quality against medium on 50 of your own briefs.
5. Batch for anything that waits. Half price on every 2.1 rate. 4K drops to $0.0567.
6. Grounding cap. 5,000 free searches a month, then $14 per 1,000. One request can run several. Turn it off where you do not need it.
7. Spend cap. A project spend cap in AI Studio on the creative project, with a named owner.
8. Migration notice. Nano Banana 2 is deprecated with no shutdown date. Ask for written notice before shutdown.
9. Pass-through for resellers. If a creative tool runs on Google image models, get in writing which model and resolution each credit buys, and whether the rate cut lowers your price.

Who owns which line

• Procurement: price per image in writing, token counts per resolution, migration notice, reseller pass-through.
• Marketing ops and creative ops: thinking level, Batch routing, and the 50-brief quality test.
• GTM engineering: token logging per request, grounding settings, and the 4K re-price.
• Revenue Finance: the spend cap and the monthly creative budget.

Soft opens

• All prices are Gemini Developer API list. Google says Gemini Enterprise Agent Platform prices may differ.
• The request shape above is an example. Google has not published typical input or thinking token counts per image.
• Quality claims on the model card are vendor-run. Test on your own briefs and brand assets.
• Google's pricing page changed once on launch day. Check it again before you sign.

Sources

1. Google, Gemini Developer API pricing (last updated Oct 7 2026): ai.google.dev/gemini-api/docs/pricing
2. Google, Gemini API release notes (Oct 6 2026): ai.google.dev/gemini-api/docs/changelog
3. Google, Gemini Nano Banana 2.1 model page: ai.google.dev
4. Kingy AI, Nano Banana 2.1: Specs, Pricing and Benchmarks (updated Oct 7 2026): kingy.ai
5. Scoy, AI News Roundup for October 7 (Oct 7 2026): scoy.ai

New Order Form decoders land in the weekly issue first. Subscribe.