GPT Image 2.5 reference image cost: Runway credit vs Sume tokens

On Runway, a GPT Image 2.5 reference image costs 1 credit and is charged once per request, not per output image. On Sume there is no flat reference fee: reference images count as input image tokens, billed at $8 per million, inside the endpoint price.

How does Runway count reference images?

The page gives the formula outputCount × perImage + referenceCount × 1. Its example: four 1K medium images with two references is 4 × 6 + 2 = 26 credits. Reference images are charged once per request.

How does Sume count them?

Both openai/gpt-image-2.5 and openai/gpt-image-2.5-sunburst support text-to-image and up to 16 image references. The token rates in the docs are $30 per million output image tokens, $8 per million input image tokens and $5 per million input text tokens, before Sume pricing is applied. Input token counts are estimates, and the total is rounded up to $0.0001.

Which price do I actually pay on Sume?

The endpoint pricing line. It is the amount charged to your wallet, with Sume's margin already applied. Read it from the catalog rather than recomputing from token rates, because token counts are estimates.

Does the number of outputs change the reference cost?

On Runway, no: references are added once. On Sume the docs do not publish a per-reference flat amount, so test one request with your real references and read the quoted price. For limits and file handling, see GPT Image 2.5 reference images.