GPT Image 2.5 with 16 references costs 5 times a text-only call

On ChatGPT Image 2.5 every reference you attach adds an estimated input-token charge, so a high-quality 1024x1024 image costs $0.0659 with no references, $0.0835 with one and $0.347 with the maximum 16, about five times the text-only price. The Image API docs price input image tokens at $8 per million and note that input token counts are estimates. Sume's reservation includes that estimate, and the response reports what was billed.

If you attach 16 images because the limit allows it, you pay for them whether or not the prompt needs them.

What does each reference add?

At high quality and 1024x1024, Sume's estimate adds about $0.0175 of billable cost per reference, from roughly 1,756 estimated input tokens each. The curve is close to a straight line.

How many references should I send?

Send the ones the output must obey, and no more. An avatar pose needs one. A product in a scene needs the packshot and maybe a scene reference. A try-on needs a person and a garment. Reserve 16 for a real multi-item composite, such as a basket built from six item photos.

Check the effect on your own prompts: render the same edit with 2 references and with 6, compare, and look at usage.cost on each response.

Does quality change the input charge?

Yes, in Sume's estimate the input-token count rises with quality: one reference at medium is about $0.0209 and at low about $0.0094, so on cheap tiers the reference charge is a large share of the bill. Prices here are list x 1.25 in dollars per image before whole-cent rounding; the catalog's billable formula reads "list x 1.25, ceil usd cents", so confirm the first charge in usage.cost and budget from the invoice, not from the sum.