GPT Image 2.5 batch of 4 at 1920x1080: price by quality on Sume

Four GPT Image 2.5 images at 1920x1080 in one call cost 3 cents at low, 6 at medium, 20 at high, 36 at xhigh and 80 at max on Sume.

3 min readSume
All posts

Four GPT Image 2.5 images at 1920x1080 cost 3 cents at low quality, 6 at medium, 20 at high, 36 at xhigh and 80 at max in one Sume call. At low and medium that is cheaper than four single calls.

How GPT Image 2.5 is priced

The price comes from output tokens at $30 per 1M image tokens plus input tokens, rounded up to $0.0001, then Sume adds 1.25 and rounds the job up to the cent. num_images accepts 1 to 4. Auto quality reserves the maximum, per the image models docs, so name a quality when cost matters.

GPT Image 2.5 at 1920x1080, billed per job (read 2026-10-06)
Quality1 image4 images, one call4 single calls
low$0.01$0.03$0.04
medium$0.02$0.06$0.08
high$0.05$0.20$0.20
xhigh$0.09$0.36$0.36
max$0.20$0.80$0.80

Reading it

The saving is rounding. A single low image is 0.56 cents, billed as 1; four are 2.2 cents, billed as 3. From high up, the per-image cost is already large enough that rounding no longer helps, and the four-image price equals four single prices.

The tiers are far apart: for the four-image batch, max at $0.80 is 4 times high at $0.20 and 27 times low at $0.03. Use low or medium to explore a prompt and keep max for a final.

Plan a batch

For 100 prompts at 4 variants each at medium quality, that is 100 calls at 6 cents, $6.00 in all. At 4K the numbers change; see the 4K price post and, for per-thousand math, the sync vs batch comparison.

  • Always name a quality.
  • Batching helps at low and medium only.
  • Prices above are for 1920x1080.

Why auto is risky for budgets

The docs state that auto quality reserves max, so a call that leaves quality unset is reserved at the maximum tier. At 1920x1080 that is 20 cents for one image, or 80 for four, against 2 and 6 at medium. Name the quality explicitly in every request that you want to cost-control.

If you do not know which quality a prompt needs, start at low. It is one cent per image at this size, so you can try a dozen prompts for pennies. Move up a tier only for the prompts that show promise, and use xhigh or max rarely, since the max tier is 10 times the cost of a medium image at this size.

How these numbers were produced

The figures are Sume's billed-amount estimates for gpt-image-2.5 on 2026-10-06, at 1920x1080 with one and four images at five quality tiers. The tables show whole cents.

These are estimates of the billed amount for the stated settings. Anything not in the table, such as a different frame shape, a reference input or a changed quality tier, can change the number, so run the estimate for your own request before a large batch. Where this page and the public API pricing list disagree, the pricing page and the live quote win.

Sources

Related posts

More in Pricing

All Pricing posts

Written by Sume