GPT Image 2.5 low 1K to high 4K is a 9x step: $0.02475 to $0.2225

GPT Image 2.5 on Sume goes from $0.02475 per image at low quality and 1K to $0.2225 at high quality and 4K. The ratio is 0.2225 / 0.02475 = 8.99. The middle tier, medium at 2K, is $0.055625, which is 2.25 times the bottom and a quarter of the top (0.055625 / 0.2225 = 0.25). The price list ties quality and size together, so these three points are the ones to budget on.

If you omit quality, the docs say the default is high, so an unconfigured call lands near the top of the ladder.

The three list points

Each Sume price is the provider list price times 1.25, and billed amounts round up to the cent. The provider column below is the Sume price divided by 1.25.

What 100 images cost at each point

One hundred images are $2.475 at low 1K, $5.5625 at medium 2K and $22.25 at high 4K. Mixed workloads are where it pays to plan: 80 low drafts and 20 high finals cost 80 x $0.02475 + 20 x $0.2225 = $1.98 + $4.45 = $6.43, against $22.25 for all 100 at the default high 4K.

Because quality and size are separate fields, check the response's usage.cost for any combination not on the list. The docs also note that auto quality reserves the max tier, which holds more balance than a high request does, so set the field in code.

A rule of thumb

Draft at low, 1K. Move to medium, 2K when text or fine detail matters. Reserve high, 4K for images that will be printed or cropped hard. Each step up costs 2.25 times and then 4 times the previous one (0.2225 / 0.055625 = 4), so the last step needs a reason.

Budgeting a month

A team that makes 1,500 images a month can use the ladder to set an allowance. If 1,200 are low drafts (1,200 x $0.02475 = $29.70), 250 are medium at 2K (250 x $0.055625 = $13.90625) and 50 are high at 4K (50 x $0.2225 = $11.125), the month costs $54.73 before cent rounding. The same 1,500 images at the default high 4K would cost 1,500 x $0.2225 = $333.75.

The point is not that high 4K is poor value. It is that the default sits at the top of the ladder, so a team that never sets quality pays the top price for every draft.

What the price list does not say

The three points above are the ones the Sume price list uses for GPT Image 2.5 in the shared pricing data. The Image API docs also give token-based figures for xhigh and max at 1024 x 1024 ($0.09366 and $0.21072 before Sume pricing). Anything else, such as medium at 1K or high at 2K, should be measured with a real request, because the docs say token counts in usage are 0 in v1 and cost is the field to read.