gpt-image-2.5 xhigh and max at 2048 and 4K: up to $0.54 an image

gpt-image-2.5 max costs $0.5353 at 2048x2048 and $0.5004 at 3840x2160 on Sume, and xhigh costs $0.2379 and $0.2224. Those two tiers are where a batch stops being cheap: a hundred max images at 2048 is $53.525.

The Image API lists xhigh and max next to low, medium and high, and says high is the default when you omit quality. Nothing forces you above high.

Top tiers by size

Same model, explicit image_size, Sume price per image.

A guard before the expensive call

Sume reserves the price at submit and returns 402 insufficient_credits if the balance cannot cover it, but your own ceiling is cheaper than a failed batch. Refuse the tier in code unless the job is flagged as a final.

ALLOWED = {"draft": "low", "review": "medium", "final": "high"}

def quality_for(stage: str, hero: bool = False) -> str:
    if hero and stage == "final":
        return "xhigh"
    return ALLOWED[stage]

print(quality_for("draft"), quality_for("final"), quality_for("final", hero=True))

Where the money goes

At 1024x1024 the max tier is 4.0x high. At 3840x2160 it is 4.0x. Moving a 4K image from high to max adds $0.3753 per image.

Run a ten-image trial at each tier on your real prompts and compare. If you cannot tell the difference at the size you publish, do not pay for it.

Checklist

Before you raise a tier, run through these.

• Pin quality and image_size explicitly. Leaving quality on auto reserves max.
• Try high at the final size first; it is the default and the cheapest tier that most listings need.
• Keep xhigh and max for the one image per product that leads the page.
• Read GET /v1/images/models/{model_id}/endpoints for the live pricing line before a large run: prices are catalog data and can change.