GPT Image 2.5 quality auto reserves max: set quality yourself on Sume
On Sume, GPT Image 2.5 quality auto reserves max. At 1024px the max output estimate is $0.2634 with margin versus a $0.02475 low row. Set quality explicitly.
On Sume, quality: auto for GPT Image 2.5 reserves the max quality estimate, so leaving quality unset or on auto holds the most money for the job. Set quality explicitly: the low 1K row is $0.02475, while the max output estimate alone at 1024 x 1024 is $0.21072 before Sume's margin, which is $0.2634 after multiplying by 1.25.
What the docs say
Sume's Image API docs say GPT Image 2.5 accepts quality: auto|low|medium|high|xhigh|max, and that if you omit quality the default is high. They also say that auto quality reserves max, and that auto size, and named presets without a verified GPT-specific pixel mapping, reserve the upper bound of output tokens.
Two settings are in play. Omitting quality gives high. Sending quality: "auto" reserves max. They are not the same, so a client that forwards a literal auto is reserving more than one that sends nothing.
The numbers
The docs price output tokens at $30 per million. At 1024 x 1024 they give two output estimates, which convert to tokens by dividing by $0.00003.
| Item | Amount | Arithmetic |
|---|---|---|
| xhigh output | $0.09366 | 3,122 tokens x $30/M |
| max output | $0.21072 | 7,024 tokens x $30/M |
| max output with 1.25 margin | $0.2634 | 0.21072 x 1.25 |
| Catalog row, low 1K | $0.02475 | fixed row |
| Catalog row, medium 2K | $0.055625 | fixed row |
| Catalog row, high 4K | $0.2225 | fixed row |
What to do
Reservation is not the final charge: Sume reserves an estimate at submit, then captures the billable amount for the completed generation. But the held amount limits how many jobs you can have in flight on a small wallet. Ten auto jobs hold about 10 x $0.2634 = $2.634 against 10 x $0.02475 = $0.2475 for ten low jobs.
Pick the tier by purpose. Use low for drafts, medium for working copies, high for finals, and keep xhigh and max for dense text or print work. Check the live pricing lines on GET /v1/images/models for the model you call, because those are the amounts the wallet is charged.
A client-side guard
If you build an app on top of Sume, validate quality before you send it and default to low or medium in drafts. Sending auto should be a deliberate choice, since it reserves the max estimate. Log the usage.cost from each response so you can see what each tier actually billed.
Keep in mind the same rule for size: auto size reserves the upper bound of output tokens, so pass a named size when you know the shape you want.
Sources
Related posts
More in Models
- Grok Imagine Video 1.5 on Sume: image-in only, silent, $0.19 for 15 s
Grok Imagine Video 1.5 on Sume needs a start image, makes silent 480p or 720p clips of 4 to 15 seconds, and a 15-second clip is 15 x $0.0125 = $0.1875.
- Haiku 5.5 cache write vs read: break-even on a Sume tool catalog
Haiku 5.5 charges $0.125 per million to write cache and $0.01 to read it. For a 12,000-token Sume tool list the cache pays back on the first reuse.
- HappyHorse 1.1 image-to-video: 300 px, 20 MB and ratio 1:2.5 to 2.5:1
Alibaba's HappyHorse 1.1 image-to-video accepts JPEG, PNG or WEBP of at least 300 px, up to 20 MB. Prepare a first frame, or use an Omni image_url on Sume.
- HappyHorse 1.1 API watermark defaults to on: how to turn it off
Alibaba's HappyHorse 1.1 image-to-video API adds a Happy Horse text mark bottom-right unless you set watermark to false. What to check before delivery.
Written by Sume