Which Sume image models accept quality, and which reject it with a 400

Only ChatGPT Image 2.5, ChatGPT Image 2, Ideogram V3 and Ideogram 4.5 list a quality value in Sume's image catalog. On every other row the catalog's quality list is empty, and since Sume rejects parameters that a model does not list with 400 unsupported_parameter (it never drops them silently), a stray quality field fails the request.

The rows

Price effects

Quality is the main price lever where it exists. Ideogram 4.5 bills $0.0375, $0.075 or $0.275 for low, medium and high. ChatGPT Image 2.5 at 1024 square bills $0.007375 at low, $0.0165 at medium, $0.065875 at high, $0.1171 at xhigh and $0.2635 at max (output tokens only, list x 1.25). Setting auto on 2.5 reserves the max price before the job runs, so pin the value.

A client that cannot break

Do not hard-code which models take quality. Read the descriptor and only add the field when the row lists it:

import os

import requests

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
rows = requests.get("https://api.sume.com/v1/images/models", headers=H, timeout=30).json()["data"]
for row in rows:
    quality = row["supported_parameters"].get("quality")
    values = quality["values"] if quality else "none (do not send quality)"
    print(row["id"], values)

Choosing a value

For a first pass on rows with a quality knob, use the lowest tier that shows your text legibly, then raise it only where the output fails. On Ideogram 4.5 the steps are $0.0375, $0.075 and $0.275 billed; on ChatGPT Image 2.5 at 1024 they are $0.007375, $0.0165 and $0.065875 for low, medium and high, and higher still for xhigh and max.

For rows with no knob, the lever is the model and the prompt. A client that reads the descriptors can offer the right controls for each model without a hard-coded table, which is also what keeps it working when Sume adds a row.

Notes

• If you port a script that always sends quality: "high", route it through the descriptor check above or it will 400 on Nano Banana, Seedream and Flux.
• OpenAI says earlier GPT Image models only support up to high (OpenAI guide, read 2026-10-07).
• The Image API page explains the capability descriptors.