How much do 1,000 AI images cost on the Sume image API?

A thousand images on the Sume image API cost anywhere from about $5 on Soul to about $264 on the ChatGPT Image 2 catalog row, at the per-image list price each model publishes in GET /v1/images/models/{id}/endpoints. Most of the catalog sits between $25 and $100 per thousand. This post multiplies the catalog's own pricing line by 1,000 so you can budget a batch before you write any code.

The prices below are the cost_usd on each model's output_image pricing line, read 2026-10-03. Sume's Image API page says endpoint pricing is the amount charged to your wallet, with Sume's margin already applied, so cost_usd × n is what a call costs. A failed or cancelled generation is not billed.

Price per image and per 1,000 images, cheapest first

Sorted by the per-image list price. Soul is the cheapest row and ChatGPT Image 2 the most expensive; the ranking says nothing about quality, only cost.

Four rows where the list price is only a default

The catalog publishes one number per model, and four models vary around it:

• Nano Banana 2 and Nano Banana Pro: the listed price is the 1K default; the catalog description says the reserve scales for other tiers (512, 2K, 4K on Nano Banana 2; 4K on Pro).
• ChatGPT Image 2.5 (both ids): the listed price is high quality at 1024×1024 output only. quality runs from low to max, custom sizes change the output size, and reference images add input cost, so budget from a real quote rather than the list line.
• ChatGPT Image 2: described as a high/1024 display rate; the actual charge follows quality and size.
• Ideogram 4.5: the listed price is the default medium quality; low and high are priced differently.

How to budget a batch

Multiply the per-image price by n and by the number of calls. For a batch where a pick rate matters, price the draft model and the final model separately: for example 1,000 drafts on a $0.025 model is $25, and re-rendering the best 100 on a $0.075 model adds $7.50. Anything that returns an error is not billed, so retries on failures do not add to the total.

To see the same numbers from your own account instead of this table, call the catalog directly:

curl https://api.sume.com/v1/images/models/qwen/qwen-image/endpoints \
  -H "Authorization: Bearer $SUME_API_KEY"

Where the numbers come from

Catalog prices can change. The table is a dated snapshot; the endpoint is the source of truth. sume/auto is not in the catalog (the docs say Sume does not disclose which family ran), so price Auto runs from the usage on the response rather than from this table.