Qwen Image vs Grok Imagine: same 2.5 cents on Sume, 4 vs 1 per request

Qwen Image and Grok Imagine are tied on price at $0.025 per image on Sume ($25.00 per 1,000), so the choice is not about cost. It is about how many images you get per request, how the output looks on your prompts, and which one handles the aspect ratios you need.

The one difference the catalog states plainly is n. Qwen Image lists up to four images per request; Grok Imagine lists one. That decides how many calls a batch needs.

The tie

Both rows bill provider list ($0.02) times 1.25, per the Image API docs, and both are edit-capable, with a reference range of 0 to 10 in the catalog. At 100 images each costs $2.50; at 10,000 each costs $250.00.

What the request count changes

A request has overhead beyond its price: a network round trip, a job record, a result to store. A batch of 1,000 needs 250 requests on Qwen Image and 1,000 on Grok Imagine. The money is the same, but the second run has four times as many things to log, retry and rate-limit.

On the other hand, a four-image request that fails costs you four missing images at once, and you retry the lot. Grok's smaller unit gives finer retries. Neither is wrong; pick the one your pipeline handles well.

How to decide on the content

Send the same twenty prompts to each. That costs $0.50 per model, $1.00 for the pair. Check the aspect ratios in each capability descriptor first: the Sume docs note that Grok does not include 4:5, so a portrait Instagram post needs a different row or a crop.

If you need a seed to repeat a result, neither helps: Sume rejects seed on every model with 400 unsupported_parameter.

Edits and references

Both rows accept reference images, so both can do product shots from a supplied photo. The practical difference shows up in what you ask for: with Qwen Image you can ask for four takes on the same edit in one call and pick one; with Grok Imagine you send four calls with slightly different wording, which also gives you four different prompts to compare.

For a 25-edit trial, Qwen Image needs 7 requests (six of four images and one of one) and Grok needs 25; both bill 25 x $0.025 = $0.62.

A note on volume

At 10,000 images the tie is worth $250.00 to each model, so the choice between them will not move your budget. What will move it is the retry rate. If one model needs two attempts per usable image and the other needs one, the first costs twice as much in practice. Count usable images, not generated ones.

Summary of the trade

• Choose Qwen Image when you want several variants of one prompt in a single call.
• Choose Grok Imagine when you want one image per call and the output matches your taste.
• Choose by test results, since the prices cannot separate them.
• Read each model's descriptors from GET /v1/images/models before you code against them.