One prompt on every Sume image model: about $1.36 in total

Sending one prompt to every image row in the Sume catalog costs about $1.36 at default settings, one image each. That is the sum of 19 billed per-image prices from the repository pricing tables on 2026-10-08: provider list times 1.25. ChatGPT Image 2 alone adds $0.264 of that, and six rows bill at or under four cents.

A one-prompt sweep is the quickest honest model comparison. It also doubles as a smoke test after a launch week such as this one, when Nano Banana 2.1 became generally available and the retired Nano Banana 2 id began running as 2.1.

The bill, row by row

Prices are per image, before any cent rounding on the total. Some rows vary by setting; this table uses the default.

Run the sweep

• Read GET /v1/images/models and loop over data[].id; the live list may be shorter than this table.
• Send the same prompt with mode: "async" so slow rows do not stall the loop, then read each result_url.
• Store usage.cost per row next to the image file.
• Skip rows that are text-to-image only if your prompt includes a reference.

Make the sweep repeatable

Save the prompt, the ratio and the date with the images. Next month, run the same sweep again and compare billed price per row; a changed number tells you the catalog moved, and a changed image tells you the model moved. Keep the sweep to one image per row, since n multiplies the cost.

If you have a reference photo, run a second sweep for the edit-capable rows only, which excludes Soul, both Imagen rows, Recraft V4 and Qwen Image Max. That second sweep costs about 25 cents less than the full one.

How to read the results

Put the images in one grid in the order of price. If your prompt is a product shot, ask three colleagues to pick the best two without telling them the models. The cheapest row that makes the shortlist is your draft model; the best is your final.

Check ratios before you compare: not every row lists every shape, and a missing ratio will fail with 400 instead of falling back. Use 1:1 for the first sweep, since all rows list it.

What Sume does not do

The total above assumes the catalog is fully available; Sume hides a row when its platform provider is not configured. It also does not add a discount for a sweep. GPT Image rows price by size and quality, so a 4K max-quality request costs much more than the default shown.