Does gpt-image-2.5 cost more at 4K? Prices from 1024 to 3840x2160

A 3840x2160 gpt-image-2.5 image at high costs $0.1251 on Sume, about 1.9x a 1024x1024 image at $0.0659. Size matters less than you would guess: a 4K frame has eight times the pixels of the square but costs under two times as much.

The size rules come from the Image API: image_size takes custom pixels where both edges are multiples of 16, the longest edge is at most 3840, the aspect ratio is at most 3:1, and the total is 655,360 to 8,294,400 pixels.

Price by size

Same model, same prompt class, only image_size changes.

Reading the table

The portrait and landscape 1.5-megapixel sizes are cheaper than the square: $0.0515 against $0.0659 at high. The 2048x2048 square is $0.1339. If your layout is a banner or a story, ask for that shape instead of cropping a square.

Invalid sizes fail early

Check a size before you send it; the rules above are cheap to test in code.

def valid_size(w: int, h: int) -> bool:
    px = w * h
    return (
        w % 16 == 0 and h % 16 == 0
        and max(w, h) <= 3840
        and max(w, h) / min(w, h) <= 3
        and 655_360 <= px <= 8_294_400
    )

for w, h in [(1024, 1024), (3840, 2160), (1920, 1080), (4096, 2160)]:
    print(w, h, valid_size(w, h))

Note on 1920x1080

1080 is not a multiple of 16, so 1920x1080 fails the check above; 1920x1088 passes. Request that and crop eight rows if you need exact HD.

Picking a size

Choose the size from where the image will be shown, then price it from the table.

• A square 1024 at high is $0.0659; a 1536x1024 landscape is $0.0515.
• A 3:1 banner at 3072x1024 is $0.0371 at high, the widest shape the size rules allow.
• Remember the 3:1 ceiling: anything wider needs a crop from a valid size.
• Leave image_size or quality on auto and Sume reserves the upper bound (auto quality reserves max), so name both for a predictable hold.
• Read GET /v1/images/models/{model_id}/endpoints for the live pricing line before a large run: prices are catalog data and can change.