Which AI image models cost under 4 cents on the Sume image API?

Six rows in the Sume image catalog list under four cents per image, read 2026-10-03: Soul at $0.005, Grok Imagine, Qwen Image and Imagen 4 Fast at $0.025 each, Seedream 4.0 at $0.0325 and Flux 2 Pro at $0.0375. Together they cover drafts, thumbnails and bulk variants where you plan to re-render only the winners on a more expensive model.

Prices are the output_image pricing line from GET /v1/images/models/{id}/endpoints; the Image API page says that line is what your wallet is charged and that failed generations are not billed.

The cheap rows and what each accepts

Flux 2 Pro is the sixth row and edits with up to 10 references at $0.0375. The other five are in the table.

Limits that matter for a draft pass

Each cheap row has a catch you should know before you build on it.

• Soul takes num_images of 1 or 4 per the catalog description and resolution of 720p or 1080p; it is text-to-image only and lists no output_format.
• Grok Imagine returns one image per call (n range 1 to 1), so a batch of four is four calls.
• Imagen 4 Fast is text-to-image only and lists five aspect ratios (1:1, 16:9, 9:16, 4:3, 3:4).
• Qwen Image, Seedream 4.0 and Flux 2 Pro take references; the other three do not.

A draft-then-final budget

A concrete plan: 200 drafts on Qwen Image at $0.025 is $5.00. Pick the 20 best prompts and render each once on a pricier row, for example Ideogram V3 at $0.075 for $1.50. Total $6.50. Drafts only help when the cheap model's output predicts the expensive one's, which is a property you have to check on your own prompts; Sume does not state that it holds.

Remember that a call that exceeds the 30-second wait returns a 202 job instead of images, so a batch script should handle both status codes.

import os
import requests

resp = requests.post(
    "https://api.sume.com/v1/images",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json={
        "model": "qwen/qwen-image",
        "prompt": "flat-lay of a leather notebook and a pen",
        "n": 4,
    },
    timeout=60,
)
resp.raise_for_status()
if resp.status_code == 202:
    print("still running:", resp.json()["data"]["status_url"])
else:
    for image in resp.json()["data"]:
        print(image["url"])