Nano Banana 2.1 thought images: free at Google, one price on Sume

Nano Banana 2.1 draws interim "thought images" while it reasons about a prompt, and Google's guide says they are visible in the backend but not charged. You pay for the final image. On Sume the bill is simpler still: one flat price per finished image at the tier you chose.

What does Google charge for?

Google's guide describes the thinking mode in a single sentence, and its pricing page lists separate rates for the pieces. Taken together:

Does thinking change the image price?

Not the image price: a 1K image is still $0.0336 and a 2K image $0.0504 at Google's standard rate. What can move is the text and thinking output on the call, which is billed at the text rate. Google's page prices text and thinking output at $7.50 per 1M tokens, so keep prompts short if you are watching the last cent.

How does Sume bill it?

Sume's image route bills a completed image at a flat per-image price and nothing else. Token counts in the response are always 0, the images docs describe no field for thought images, and a generation that fails is not charged. The billed prices are $0.10, $0.15 and $0.20 for 1K, 2K and 4K, which is the list price times 1.25.

If you need to see or control the model's reasoning, call Google's API directly; the Sume images docs do not describe a way to read it.

• One call, one image, one flat price.
• n above 1 multiplies the price; thinking does not.
• Check GET /v1/images/models/{id}/endpoints and one test call for the current numbers.

Why does this matter for a budget?

A flat price is easy to forecast: 200 product images at 2K is 200 times $0.15, or $30. A token-metered price needs your prompt lengths and thinking output. Read more in the image models docs.