Gemini 3 Pro Image: $0.24 at 4K vs $0.134 at 1K/2K, what to budget

Google's Gemini API page lists Gemini 3 Pro Image (Nano Banana Pro) at $0.134 per image for 1K or 2K output and $0.24 per image for 4K, so a 4K image costs about 1.8 times a 2K one. Budget 100 finals at 2K as $13.40 and at 4K as $24.00 on Google's direct rate, then read the Sume endpoint price line, which is a different number, before you submit.

What does Google list for Gemini 3 Pro Image?

The page shows token pricing of $2.00 per million input tokens and $120.00 per million output tokens, with per-image equivalents: $0.134 for a 1K or 2K image and $0.24 for a 4K image. Note that 1K and 2K share one price, so the cheapest way to get more pixels from this model is to ask for 2K rather than 1K.

Why is the 4K step worth checking before a batch?

The 4K tier is $0.106 more per image than 2K. On a 500-image catalogue refresh, that is $53 you spend only if the 4K file is used at 4K. If the destination crops or downsizes the image, generate at 2K, which Google prices the same as 1K. If you need 4K for print or large displays, budget it explicitly.

How does this map to a Sume budget?

The Google figure is the vendor's direct rate, not a Sume price. On Sume the number you pay is the cost_usd on the model's endpoint record, with Sume's margin already applied, multiplied by n. For listed models Sume's rule is provider list price times 1.25, reserved at submit. Provider list prices on a hosting provider can differ from Google's direct page, so do not assume Google's $0.134 times 1.25.

The images API accepts a resolution tier of 512, 1K, 2K or 4K. Which tiers a given model supports is in its endpoint record, so check it first; do not assume every tier is available on every model.

curl "https://api.sume.com/v1/images/models/$MODEL_ID/endpoints" \
  -H "Authorization: Bearer $SUME_API_KEY"
# MODEL_ID: the Nano Banana Pro id from GET /v1/images/models
# check supported_parameters for resolution values
# then read pricing[].cost_usd and multiply by n

What can go wrong with the wallet?

Sume reserves the estimated cost when a request is accepted. If the wallet cannot cover the reserve, submit fails with 402 insufficient_credits before provider work starts, as described in Generation admission. A 4K batch reserves more per image, so a balance that covers 100 images at 2K may not cover 100 at 4K.

Top-ups are a dashboard operation: the public API exposes GET /v1/balance and GET /v1/usage reads, not a top-up endpoint. Check the balance first, as Billing and credits describes.