Nano Banana 2.1 prices: Google's 1K and 4K list vs Sume's cost

Google's own list price for Nano Banana 2.1 is $0.0336 for a standard 1K image and $0.0756 at 4K, which is about half the $0.067 it reports for the previous version. Sume does not bill Google's number. Sume serves the model as google/nano-banana-2.1 with its own per-image price, and you read that price from the catalog before you spend anything.

What Google announced on October 6

Nano Banana 2.1 launched on 2026-10-06 across the Gemini app, Search AI Mode and developer tools. The launch coverage lists better visual design, sharper mask-based editing and better subject consistency across repeated edits. Treat the benchmark claim (1,050 Elo against 990 for Nano Banana 2) as Google's own, because independent verification was not available at launch.

Why Google's list price is not your Sume price

A vendor list price is what the vendor charges its API customers. On Sume, each model has one endpoint, and the endpoint record carries the billable line that Sume charges your wallet. Sume's image docs say those lines already include the Sume margin, so you pay cost_usd × n. Do not multiply Google's figure by anything yourself.

Read the Sume price in two calls

Ask for the endpoint record of the model. The pricing array is the number to budget with, and the supported_parameters block shows which resolution tiers and aspect ratios the model accepts.

curl "https://api.sume.com/v1/images/models/google/nano-banana-2.1/endpoints" \
  -H "Authorization: Bearer $SUME_API_KEY"

curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"google/nano-banana-2.1","prompt":"studio product photo of a ceramic mug on a linen cloth","resolution":"1K"}'

Check usage.cost after the first image

The response usage.cost field is the USD amount billed to your wallet. Token counts in usage are always 0 on Sume, because image models are metered per image, so do not compute spend from tokens. If you ask for 4K and the render takes longer than the 30-second wait, the call returns 202 with a job envelope, and you read the images from GET /v1/jobs/{id}/result.

The retired id google/nano-banana-2 still works and runs as 2.1, and the job stores the 2.1 id. If you pinned the older id for cost reasons, re-read the endpoint pricing instead of assuming the old figure.

What this does not tell you

Google's batch discount is a Google API feature. The Sume docs do not describe a batch discount for images, so do not plan around one. Whether the 4K tier is available for this model on Sume is shown by the resolution descriptor in the catalog row, and a tier the model does not list is rejected with 400 unsupported_parameter.

Three checks before moving a batch to 2.1:

• Read the endpoint record and note the cost_usd line for the tier you will use.
• Run three images and compare the summed usage.cost against your estimate.
• Set resolution explicitly so the reserve matches what you intend to generate.