High to max on 100 hero images: GPT Image 2.5 adds $20, xhigh adds $5
Upgrade 100 GPT Image 2.5 hero images from high to xhigh or max: the extra cost on Sume is $5 or $20 at 1024-class size. The arithmetic is shown.

Re-rendering 100 hero images on GPT Image 2.5 at xhigh instead of high costs $5 more, and at max it costs $20 more, for 1024-class output on Sume. High is 7 cents per image, xhigh about 12 and max about 27, so the totals are $7, $12 and $27 for 100 images.
The upgrade math
High is the default when you omit quality. The Image API docs give 1024x1024 output estimates of $0.09366 (xhigh) and $0.21072 (max), before input tokens and Sume pricing. Multiply by 1.25 and round up to the cent to get the Sume figure. Because input tokens are added on top and reference images count as input, a reference-heavy edit can land above these numbers.
| Step | Per image before x1.25 | Billed per image | 100 images | Extra vs high |
|---|---|---|---|---|
| high | $0.0527 | 7 cents | $7 | - |
| xhigh | $0.09366 | 12 cents | $12 | +$5 |
| max | $0.21072 | 27 cents | $27 | +$20 |
Where it is worth paying
Only GPT Image 2.5 and its Sunburst twin take xhigh and max. ChatGPT Image 2 does not take them, because its catalog lists only low, medium and high, and Sume rejects a value a model does not list. So the upgrade is a model choice as well as a setting: if you are on ChatGPT Image 2 at 27 cents for high, GPT Image 2.5 at max costs the same 27 cents.
That tie is a useful fact for a migration. At equal price you get the xhigh and max tiers, the 16-image reference limit and a transparent background option. The docs do not publish a quality comparison, so look at a pair of renders before you decide.
- Use xhigh when high is nearly right and a $5 per 100 bump is fine.
- Use max for a short list of final images: 20 of them is $5.40.
- Do not use auto quality for budgets, because auto reserves the max amount.
Order of operations
Draft at low, shortlist, then upgrade only the shortlist. Keep the prompt and references identical between runs so you compare the tier and nothing else.
Budget the whole campaign
Suppose a launch needs 100 hero images and you render each three times before choosing. At high that is 300 images and $21. Do the three tries at low (300 x 1 cent = $3), choose one per hero, then render the 100 winners at max ($27): a total of $30, against $81 if every try were rendered at max. The drafts are cheap enough that you can generate them freely and spend the premium only where the eye needs it.
Sources
Related posts
More in Models
- How many aspect ratios does each Sume image model list? Oct 2026
Nano Banana 2.1 lists 15 aspect ratio values, Imagen 5. A count for every Sume image row, and why the number decides banners, stories and 4:5 posts.
- How many aspect ratios each Sume image model lists, vs Hy's 14
Hy Image 3.5 Preview lists 14 aspect ratios. On Sume, Nano Banana 2.1 and Ideogram list 15, Grok and Flux 13, GPT Image 2.5 only 9. Full counts by row.
- HunyuanVideo VRAM: 45 GB at 544p, 60 GB at 720p, 80 GB advised
The HunyuanVideo README lists 45 GB for 544x960 and 60 GB for 720x1280 at 129 frames, tested on one 80 GB GPU. FP8 saves about 10 GB. Table and hosted route.
- HunyuanVideo 720p: 1,904 s on one GPU, 338 s on eight
README latency for a 1280x720, 129-frame, 50-step clip: 1,904 s on 1 GPU and 338 s on 8. The GPU-seconds each setup burns, and what it means for hosted jobs.
Written by Sume