What 1,000 AI images cost on Sume across eight models

A thousand 1K images cost anywhere from $16.50 to $100 on Sume, depending on the model. The image docs say each endpoint pricing line already includes Sume's margin and that you pay cost_usd times the number of images, so the arithmetic is simple.

Per image and per thousand

These figures come from the repo's image router estimator, called with 1K resolution, a 1024x1024 size and medium quality where quality applies. Quality and size change the result for token-priced models, so treat the table as a starting grid, not a quote.

Billing is all or nothing

The docs state that a completed generation is billed in full, and a failed or cancelled one is not billed. A request that ends early because the client disconnected counts as failed. So the thousand-image figure is the cost of 1,000 successes, and a failed call does not add to it.

What changes the figure

Image prices on token-priced models move with quality and size. The docs for GPT-class image models say auto quality reserves the maximum, and auto size or named presets without a verified pixel mapping reserve the upper bound of output tokens. If you leave quality and size on auto, your hold is the ceiling, not the medium figure in the table.

Fixed-price models are easier to budget, because the price per image does not depend on the prompt. Either way, the endpoint's pricing lines are the number the wallet is charged, and they already include Sume's margin, so no further markup applies.

How to keep the bill down

• Prototype prompts on the cheapest model and move to a dearer one for finals.
• Ask for one image at a time while iterating; the cost is linear in n.
• Read each model's pricing lines from GET /v1/images/models before a large run, since quality, size and resolution options change them.