Seven low ChatGPT Image 2.5 images or one high: 7 cents either way

For the price of one high-quality ChatGPT Image 2.5 image you can generate up to seven low-quality ones. On Sume, a 1024-class image costs about 1 cent at low, 2 cents at medium, and 6 to 7 cents at high. So the real choice is between exploring seven options cheaply and committing to one finished one. The trap is the default: if you omit quality, ChatGPT Image 2.5 runs at high.

Cost per image and per thousand

Prices are the Fal-based catalog price times Sume's 1.25 factor, rounded up to the cent. The catalog list of $0.0527 is the high, 1024 output rate: 0.0527 x 1.25 = 0.0659, rounded up to 7 cents. Low and medium figures are the per-image values used for Sume billing at 1024-class sizes.

Quality values and the hold

quality accepts auto, low, medium, high, xhigh and max. Two things move the number. First, auto quality reserves max at admission, so pin the quality if you want a small hold on the wallet. Second, xhigh and max cost more still: at 1024 x 1024 the output list is $0.09366 for xhigh and $0.21072 for max, before input tokens and before Sume pricing. Times 1.25 that is about 12 cents and 27 cents of output cost. Input tokens add to it.

A workflow that uses both tiers

• Pass 1: run seven low-quality variants of one brief, 7 cents in total, and choose the composition.
• Pass 2: re-run the chosen prompt once at high, 6 to 7 cents, for the finished image.
• Total for the loop: about 14 cents, against 49 cents if you had run seven highs.
• If you edit from a reference, the model accepts up to 16 reference images and an optional mask_url.

Caveat on judging quality

The tier names are Fal and OpenAI settings, and the Sume docs give prices, not a visual ranking. Low may be enough for a thumbnail or a mood board, and not enough for dense small text. Test your own prompt: generate one image at each tier and look at the result at the size you will publish. Then set quality explicitly in every call through the Image API so no job runs at an accidental default.

Check it before you run it

Every figure above is a catalog list price times 1.25, rounded up to the cent, as of 2026-10-08. Catalogs change, so before a large batch, read the current model entry in the docs and recompute the one line that matters for your case. Write the arithmetic next to the job in your own notes: list rate, seconds or characters, multiplier, rounding. If the result differs from the wallet charge by more than a cent, the catalog entry has changed, and the docs page is the place to find out why.

Run one small job first. Submit a single request with the pinned model id and the settings in the tables, poll the returned polling_url until it finishes, and compare the charge with your estimate. Then scale up. Using a pinned id for the test matters, because sume/auto never names the family, so you cannot tie its charge to the row you priced.