How many images does $50 buy on Sume? Count per image model

$50 of Sume credit buys 6,779 gpt-image-2.5 low images at 1024, 10,000 Soul images or 266 Nano Banana Pro images. Full count table by model row.

3 min readSume
All posts

$50 of credit buys 10,000 Higgsfield Soul images, 6,779 gpt-image-2.5 low images at 1024, or 266 Nano Banana Pro images at 1K on Sume. The model row changes the count by a factor of 38.

Images per $50

Counts are rounded down; a request needs the price in balance at submit.

Sume price, read 2026-10-08
Model rowPer imageImages per $50
higgsfield-soul$0.00510,000
gpt-image-2.5 low, 1024$0.00746,779
gpt-image-2.5 medium, 1024$0.01653,030
qwen-image$0.0252,000
flux-2-pro$0.03751,333
seedream-5-lite$0.04371,142
gpt-image-2.5 high, 1024$0.0659759
ideogram-v4.5 medium$0.075666
nano-banana-2.1, 1K$0.10500
nano-banana-pro, 1K$0.1875266
ideogram-v4.5 high$0.275181

Plan the spend

Split the money by stage: most tries at a cheap row, a few finals at a pricey one.

# 40 dollars of drafts and 10 dollars of finals, prices from GET /v1/images/models
draft, final = 0.007375, 0.065875  # gpt-image-2.5 low / high at 1024
print(int(40 / draft), "drafts")
print(int(10 / final), "finals")

Read the live price

Prices are catalog data and can change. Read each model's pricing line from GET /v1/images/models/{model_id}/endpoints before a big run.

Reading the counts

  • A count is only a ceiling: batches that use n multiply the cost per call.
  • Failed generations are not billed, so retries do not eat the budget unless they complete.
  • gpt-image-2.5 is token-priced, so the table is for a bare text prompt. A longer prompt or an input reference adds to the per-image price, and the count falls a little.

Sources

Related posts

More in Pricing

All Pricing posts

Written by Sume