GPT Image 2.5 quality levels: low, medium, high, xhigh, max and auto

GPT Image 2.5 takes six quality values: auto, low, medium, high, xhigh and max. On Sume, leaving quality out runs the request at high; sending auto reserves funds as if you had asked for max. Higher tiers use more output tokens, so they cost more and can take longer.

Values come from Sume's Image API docs and OpenAI's Sunburst model page, which lists the same six settings. Both read 2026-09-29.

What is the difference between the quality tiers?

Sume's docs do not describe how the pictures differ visually, and this post does not either. What is documented is the effect on cost and on time: the tier changes the output token count, and Sume's docs list high quality among the slow configurations that can return a 202 job envelope. OpenAI's guide warns that complex prompts may take up to 2 minutes.

Which tier runs if I send nothing?

The auto row is the trap: the request may cost less than the reserve, but the reserve is the worst case, so your wallet needs the max amount.

How do I set the tier?

Add quality to the body. The example asks for a fast draft; move to high or above for a final.

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": "flat lay of a notebook, pen and coffee cup on a linen cloth",
    "quality": "low",
    "image_size": "1024x1024"
  }'

Do the older models have all six?

No. ChatGPT Image 2 (openai/gpt-image-2) lists only low, medium and high. A value the model does not list returns 400 unsupported_parameter; that error explained.