Grok Imagine, Qwen Image and Imagen 4 Fast all cost 2.5 cents on Sume

Grok Imagine, Qwen Image and Imagen 4 Fast all cost $0.025 billed per image on Sume ($0.02 list times 1.25), so the price does not choose between them; the request limits do. Grok Imagine returns one image per call and has the most unusual ratios, including 20:9 and 19.5:9 phone shapes. Qwen Image takes up to 4 images per call, 13 ratios and reference images. Imagen 4 Fast takes up to 4 images, only 5 ratios, and no references.

All three also reject quality and resolution because none lists them, so the same request body works across the three with only the model value changed, apart from n for Grok and input_references for Imagen.

Pick Qwen Image if you want the cheap row that does the most: it edits, it batches and it has the widest ratio list after Grok. Pick Grok for phone-shaped ratios. Pick Imagen 4 Fast only if you specifically want that model's output.

The comparison

Capabilities below are read from Sume's image catalog; the price is the billed amount. No quality or resolution tier exists on any of the three.

The n difference changes the call count

Check the numbers on a live record before you build around them: GET /v1/images/models/{id}/endpoints returns supported_parameters and pricing[].cost_usd for each model, and the n range is the max on the n descriptor.

Because Grok caps n at 1, four Grok images need four calls, while four Qwen Image images need one. The cost is the same $0.10 billed, but your code has to fan out or loop for Grok. If the pipeline renders a contact sheet of four takes per prompt, Qwen Image or Imagen 4 Fast makes it a single request.

That is also a rate-limit difference: four calls take four slots of whatever concurrency limit you have, and a single call takes one.

Phone-shaped ratios

Grok Imagine's list includes 19.5:9 and 9:19.5, matching modern phone screens, along with 20:9 and 9:20. No other model in the catalog lists these. If you want a full-bleed wallpaper or a phone-screen mockup without cropping, this is the row that offers it natively.

For ordinary 9:16 vertical video covers, all three work, since each lists 9:16.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-image",
    "prompt": "Four takes of a paper boat on a puddle, overcast light",
    "aspect_ratio": "3:2",
    "n": 4
  }'

A test plan that costs $3.75

Write 50 prompts that cover your real work: some product shots, some portraits, some text on a sign. Render each once on each of the three models at the ratio you will ship. Total: 150 images and $3.75 billed. Put the three outputs for a prompt side by side and have one person choose. Count the wins per model and look at the failures: a wrong ratio, a missing reference or a refused prompt tells you more than a score.

Because the price is equal, the winner is simply the one whose output you prefer and whose limits you can live with. You do not have to weigh cost against quality, which is rare on the image API.

What a thousand costs

A thousand images on any of the three is $25.00 billed. A 10,000-image run is $250.00. A weekly run of 300 images is $7.50 a week, or about $390 over a year. A 50-prompt test of all three, one image each, is $3.75 in total. These numbers hold only for successful images; Sume does not bill failed generations.

Keep in mind that every figure here is a billed price from Sume's catalog on the date in the table caption. Prices and limits can change, so before a large run, read the endpoint record for the exact model id and compare it with your plan. A one-minute check costs nothing, and it is the only way to be sure the number in your budget is the number on the invoice.