GPT Image 2.5 quality auto reserves max: avoid the price surprise
On Sume, GPT Image 2.5 with quality auto reserves max, and an omitted quality means high. At 1024x1024 xhigh and max cost $0.09366 and $0.21072 before input.

Do not send quality: "auto" if cost matters. Sume's Image API docs say auto quality reserves max on the GPT Image 2.5 rows, and that an omitted quality defaults to high. The provider output cost at 1024x1024 is $0.09366 for xhigh and $0.21072 for max, before input tokens and Sume pricing.
Those two numbers are the docs' figures, read 2026-10-06. Set quality to low or medium for drafts, and move up only for the image you will use.
What each setting does to the reservation
| quality value | What the docs say | Output cost at 1024x1024 |
|---|---|---|
| (omitted) | Defaults to high | Not stated in the docs |
auto | Reserves max | Reserves the max amount |
low, medium, high | Catalog-gated values | Not stated in the docs |
xhigh | Catalog-gated value | $0.09366 before input and Sume pricing |
max | Catalog-gated value | $0.21072 before input and Sume pricing |
The reservation matters on small balances
The same page says that auto size and named presets without a verified GPT-specific pixel mapping reserve the upper bound of output tokens. A wallet that is close to empty may be refused for a request that would have cost much less in the end.
The error table lists 402 insufficient_credits for that case. If you see it on a draft job, check what quality and size you sent before you top up.
A guard in front of the call
This wrapper refuses to send the risky values and fills in a cheap default, so a missing field never becomes the expensive one by accident.
import os, requests
RISKY = {None, "auto", "xhigh", "max"}
def generate(prompt: str, quality: str = "medium", allow_costly: bool = False) -> dict:
if quality in RISKY and not allow_costly:
raise ValueError(f"quality {quality!r} needs allow_costly=True")
key = os.environ.get("SUME_API_KEY")
if not key:
raise SystemExit("set SUME_API_KEY")
r = requests.post(
"https://api.sume.com/v1/images",
headers={"Authorization": f"Bearer {key}"},
json={"model": "openai/gpt-image-2.5", "quality": quality, "prompt": prompt},
timeout=90,
)
r.raise_for_status()
return r.json()
if __name__ == "__main__":
out = generate("A ceramic mug on a pale table")
print(out.get("usage", {}))Order of work
- Draft at
lowand judge the composition. - Re-run the chosen prompt at
mediumorhighfor the version you share. - Use
xhighormaxonly when the image will be enlarged or printed. - Read
usage.costfrom the response and keep it with the file.
Sources
Related posts
More in Pricing
- Holiday gift guide of 25 products: card, clip and voice cost on Sume
A holiday gift guide with 25 products, each a medium Ideogram card, a 5-second Wan 3.0 clip and a voice line, plus one render, comes to about $10.43 on Sume.
- How far does $5 go on Sume? Images, speech, transcripts, music, video
A $5 test budget buys 63 medium Ideogram 4.5 images, 111 minutes of speech or 473 transcript minutes on Sume, all-in. The count for each asset type.
- How many AI images does $50 buy? Count by image model on Sume
Eighteen Sume image-router ids, $0.005 to $0.264 an image: how many images $50 buys on each, with the 5.5% fee.
- How many AI video clips does the $40 Sume Pro plan cover?
The $40 Pro plan includes $40 of usage: 63 five-second Wan 3.0 clips at 720p, 40 Omni clips of 8 seconds, or 6 ten-second Seedance 2.5 clips.
Written by Sume