GPT Image 2 API

Nine tiers, $0.06 to $0.94 per image. That is a 15-fold spread inside one model, and this page is about where on that ladder to stand.

What the fifteen-fold difference buys

We carry GPT Image 2 as nine SKUs. Three quality levels the provider calls low, medium and high, multiplied by three resolutions, 1K, 2K and 4K. Nine prices come out of that, from six cents to ninety-four.

What actually changes inside the image between low and high is something we have not measured (and somebody else's description is not ours to repeat). How the price is built, though, is plain to see, and that is what people come here for.

Price

Two things in that grid are easy to miss.

First. At the bottom level resolution costs almost nothing, the whole run from 1K to 4K adds $0.02. At the top level the same two steps cost $0.62.

The second one pays better. gpt-image-2-low-4k costs $0.08 while gpt-image-2-medium-1k costs $0.10. A large low-quality image comes in cheaper than a small medium-quality one. When size is what you are after and fine detail is not, medium-1k is money spent for nothing.

There is a similar pair at the top. High-1k at $0.32 and medium-4k at $0.29 sit almost level on price while pulling in opposite directions, detail against size.

How to call it

The tier is picked by model id, everything else works like the rest of our image models.

curl https://api.teamtoken.store/v1/images/generations \
  -H "Authorization: Bearer $TEAMTOKEN_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2-medium-2k",
    "prompt": "open book on a wooden table, cup of coffee, side light",
    "aspect_ratio": "3:2"
  }'

The tier lives in the model name, not in a parameter

This family takes neither a quality field nor a resolution field. Both are baked into the id, and gpt-image-2-high-4k means exactly "high, 4K". Sending resolution: 4K to gpt-image-2-low-1k will not do anything, that field is not read here.

Aspect ratios, on the other hand, are richer than anywhere else in our catalog. Eight of them against five on nano-banana and three orientations on grok-image, and 3:2, 2:3 and 21:9 appear nowhere else at all.

n behaves as it does on our other image models. The field is validated and then not forwarded, so a call brings back one image. Aspect ratio does not move the price either, 21:9 costs what a square costs.

Which tier for which job

I put that table together less as a rule than as a way to stop defaulting to the top tier. The top tier earns its price when the image is singular and important. Across a hundred previews the gap between $0.06 and $0.94 turns into $6 against $94, and the choice stops being a matter of taste.

What sits next to it

If nine rungs are more than your job needs, flat-priced models are right there. grok-image at $0.02 and nano-banana-pro at $0.027, and on the latter 4K adds nothing at all. The top GPT Image 2 tier runs 35 times the price of that, which is worth paying only when you know what the extra buys.

No gallery here, but there is something to test

The least obvious claim on this page is that a large low-quality image undercuts a small medium-quality one. One scene across two tiers settles it.

The full set comes to $1.12, taken off your balance rather than run in some demo mode. If only the main question interests you, the first 2 calls at $0.18 answer it.

Questions we get