ChatGPT Image 2 vs 2.5 on Sume: $0.264 vs $0.066 per image

Answer

On Sume, ChatGPT Image 2 is listed at $0.26375 per image and ChatGPT Image 2.5 at $0.065875, so 2.5 is about four times cheaper at the catalog display rate, and it takes 16 reference images against 10. The display rate is a high-quality 1024 figure; the real charge on 2.5 depends on quality and size.

ChatGPT Image 2.5 also adds mask_url for masked edits, background: auto|transparent|opaque, and quality values of xhigh and max. ChatGPT Image 2 accepts low, medium and high only, and a request for xhigh is rejected.

How 2.5 is billed

Sume prices 2.5 from the token rates on its fal GPT Image 2.5 page (read 2026-10-04): $30 per million output image tokens, $8 per million input image tokens and $5 per million input text tokens, then adds the Sume margin. Output tokens depend on quality and size, so a low 1024x768 image estimates at about $0.005 while a high one estimates at about $0.045.

Because of that, 2.5 can cost far less than the catalog figure at low quality and much more at max in 4K, where one image estimates at about $0.50. If you omit quality, the default is high.

Which to use

• Pick 2.5 for masked edits, transparent backgrounds, many references or cheap low drafts.
• Pick 2 only if you have a prompt tuned for it and a reason to keep the older behavior. Its flat-ish catalog price is higher.
• Let sume/auto decide when you do not care. Auto continues to use Flare for the 2.5 family.

The parameter table is in the Sume Image API docs.

Before a large run

Prices and descriptors change when the catalog changes, so confirm them before you spend. Call GET /v1/images/models/{id}/endpoints for the row you plan to use and read its pricing line and supported_parameters; both come back in one response.

Then run a pilot of three to five images and read usage.cost on each response. Multiply by your planned count for a forecast you can trust. Completed generations are billed in full and failed or cancelled ones are not, so a pilot that errors costs nothing.

For big batches, use mode: "async" or mode: "webhook" with a public HTTPS webhook_url, so no request waits on the 30-second sync limit. Poll GET /v1/jobs/{id}/status and fetch GET /v1/jobs/{id}/result when the job completes.