GPT Image 2.5 max quality: $0.21 at 1024, and when to use it

Per the Sume Image API docs, a 1024 by 1024 GPT Image 2.5 output costs $0.21072 at max quality and $0.09366 at xhigh, before input tokens and before Sume pricing. Pick max only when the extra detail is worth about 2.25 times the xhigh output cost. Leaving quality out gives high.

The numbers

Flare (openai/gpt-image-2.5) and Sunburst (openai/gpt-image-2.5-sunburst) share the same Fal token rates. The output estimates use OpenAI's size and quality calculator, and input token counts are estimates. Fal rounds the total up to $0.0001.

Quality values and defaults

quality accepts auto, low, medium, high, xhigh and max. If you omit it, the default is high. auto is not a cheap setting: it reserves max, so an auto request holds the highest estimate against your balance. Auto size and named presets without a verified pixel mapping reserve the output token upper bound.

Auto model routing continues to use Flare.

A request

This sends a max request at a custom size. Both edges must be multiples of 16, with a maximum edge of 3840, an aspect ratio of at most 3:1, and 655,360 to 8,294,400 pixels. The pricing array on the model's endpoint record is the amount charged, and usage.cost on the response is the billed USD.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "A ceramic mug on a linen cloth, soft window light, product photo",
    "quality": "max",
    "image_size": "1024x1024"
  }'

When max is worth it

Use max for a final hero image, text-heavy art, or an image you will crop hard. Use high for most work and low or medium for drafts. Do a draft at a lower quality first, because iterating at max multiplies the spend. Sume pricing is added on top of the figures here, so read the live pricing lines for the amount you will pay.