What $10 buys on each Sume image model: images per wallet

Ten dollars on a Sume wallet buys 400 images on Qwen Image or Grok Imagine, 133 on Ideogram 4.5 at its default medium quality and 37 on ChatGPT Image 2. The spread between the cheapest edit model and ChatGPT Image 2 is about tenfold, so the model choice moves a small test budget more than any prompt change.

The counts below divide $10 by the billed price per image from the Sume model endpoints (read 2026-10-03) and round down. They assume one image per call and the default settings of each model.

Images per $10, all 19 models

Soul is text-to-image only and priced per image at its 720p tier; Qwen Image Max, Imagen 4 and Recraft V4 are also text-only. Everything marked Edit + text accepts reference images.

Where the defaults move the number

Four rows are defaults, not constants. ChatGPT Image 2.5 and 2 show the high-quality 1024 rate, and xhigh or max quality on 2.5 costs more. Nano Banana 2 and Pro show the 1K rate and scale with resolution. Ideogram 4.5 shows medium: low is $0.0375 and high is $0.275 on Sume, whatever the size.

A failed generation is not billed, so retries on errors do not consume the wallet; completed images you discard do.

• Cheapest edit-capable model: Qwen Image and Grok Imagine at $0.025.
• Cheapest 16-reference model: ChatGPT Image 2.5 at $0.0659.
• Largest quality step inside one model: Ideogram 4.5, from $0.0375 to $0.275.

Track the burn in code

Every 200 response carries usage.cost, the USD amount billed to your wallet for that call, so a running total needs one line per response. The same field is how you check these table values against real spend.

{
  "model": "qwen/qwen-image",
  "data": [{"url": "https://media.sume.com/img/.../0.png", "media_type": "image/png"}],
  "usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0, "cost": 0.025}
}