GPT Image 2.5 at 4:5 lists 12-14% below 1:1 at the same quality

A 4:5 portrait on GPT Image 2.5 costs less than a 1:1 square at the same quality, even though 1024x1280 has 25 percent more pixels than 1024x1024. Sume's list estimate is 12 to 14 percent lower at every tier from low to max: $0.0052 against $0.0059 at low, $0.0114 against $0.0132 at medium, $0.0453 against $0.0527 at high, before Sume's 1.25 multiplier.

This matters if you are moving a feed-image workload to GPT Image 2.5 before OpenAI's gpt-image-1 shutdown on 2026-10-23 and you want the Instagram portrait shape without paying a premium for it. OpenAI's deprecations page, read 2026-10-07, lists gpt-image-2.5-sunburst or gpt-image-2.5-flare as the replacement, and Sume serves both. The numbers below come from Sume's own price tables for openai/gpt-image-2.5, the same ones that back the pricing lines on GET /v1/images/models/{id}/endpoints.

The five-tier comparison

GPT Image 2.5 is billed from output image tokens, not from a flat per-image card. Sume's estimate follows OpenAI's size and quality calculator for 2.5, at $30 per million output image tokens, and rounds up to $0.0001. The calculator scales tokens with the shorter side of the image relative to the longer one, so a 4:5 frame lands below a square. List is the provider price; billed is list times 1.25, the standard Sume margin from the image docs.

How to ask for 4:5

Send aspect_ratio: "4:5" and a quality. GPT Image 2.5 lists 4:5 among its ratios, alongside auto, 1:1, 16:9, 9:16, 4:3, 3:4, 5:4 and 9:8. If you prefer explicit pixels, and the model is one that takes custom sizes, image_size takes {width: 1024, height: 1280}; both edges must be multiples of 16, so 1080x1350 is not a legal GPT box. Render at 1024x1280 and resize to 1080x1350 in your own pipeline if a platform wants the exact figure.

If you omit quality, Sume's default for GPT Image 2.5 is high, which is the $0.0453 row above, not the cheap one. Set the tier on purpose.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "A ceramic mug on a linen cloth, soft window light",
    "aspect_ratio": "4:5",
    "quality": "medium"
  }'

What the table leaves out

Three small additions can move a real invoice. The prompt is billed as text input at $5 per million tokens, which adds roughly $0.0001 at list for a one-sentence prompt (Sume's estimator returns $0.0133 against $0.0132 for a one-sentence prompt at medium 1:1). Reference images are billed as image input tokens, so an edit costs more than a text-only call. And quality: auto reserves the max tier at submit time, so never leave it on auto for a batch.

The sync call also has a ceiling. POST /v1/images waits up to 30 seconds and returns 200 with the image, or 202 with a job envelope you poll through GET /v1/jobs/{id}/result. Higher tiers are the likeliest to fall through to 202, so handle both status codes.

Which tier for which portrait

For feed posts and product cards, medium at 4:5 is $0.01425 billed and is a sensible baseline; compare it with the $0.0165 square. Reserve high ($0.056625) for finals with dense text, and xhigh or max for hero art that you will crop heavily. A draft pass at low costs $0.0065 billed per portrait, so ten drafts are $0.065. Ten finals at high are $0.56625, and the gap between the two is the reason to draft at low and rerender only the keepers.

If the same shot also needs a wide and a square variant, price all three before you commit: 16:9 at 1536x864 is cheaper still, at $0.0084 list for medium, but a different crop is a different composition, so only request it when the layout calls for it.

Keep in mind that every figure here is a billed price from Sume's catalog on the date in the table caption. Prices and limits can change, so before a large run, read the endpoint record for the exact model id and compare it with your plan. A one-minute check costs nothing, and it is the only way to be sure the number in your budget is the number on the invoice.