Cheapest image edit on Sume: 14 edit-capable models ranked by price

From Grok Imagine at about 2.5 cents to ChatGPT Image 2 at about 26: all 14 edit-capable Sume image models, sorted by catalog list times 1.25.

5 min readSume
All posts

Before you pick a model for an edit, rank them by price. The table is built from the catalog source at origin/main: each row is the catalog list rate per image times Sume's 1.25 margin (Sume Image API). It excludes the five text-only models, since they cannot edit.

Edit-capable models, cheapest first

These are catalog display rates before cent rounding. Two rows deserve care. Nano Banana 2 is priced at its default 1K tier, and larger tiers cost more. ChatGPT Image 2.5 is priced as high quality at 1024 pixels, but it bills by image tokens, so size and quality move the price. Read usage.cost or the endpoint pricing lines for the exact figure.

Catalog list rate and list × 1.25 per image, edit-capable Sume image models, read 2026-10-04
ModelCatalog listList × 1.25Note
grok-image$0.020$0.025n max 1
qwen-image$0.020$0.025
seedream-v4$0.026$0.0325
flux.2-pro$0.030$0.0375
seedream-5-lite$0.035$0.04375
seedream-4.5$0.040$0.050
flux.2-flex$0.050$0.0625
gpt-image-2.5 (and sunburst)$0.0527$0.0659High, 1024 px; token-billed
ideogram-v3$0.060$0.075
ideogram-v4.5$0.060$0.075Medium; low $0.03, high $0.22 list
nano-banana-2$0.080$0.1001K tier
nano-banana-pro$0.150$0.18751K tier
gpt-image-2$0.211$0.2638High, 1024 px

How to use the ranking

Rates move when catalog rows change, so treat this table as a snapshot of 2026-10-04 and refresh it from the catalog before you quote a price.

  • Draft cheap, finish dear: iterate on one of the first rows, then re-run the approved edit on the model you want to ship.
  • Cheap is not the same as faithful. A low rate says nothing about how well a model keeps your untouched pixels, so test that too.
  • A failed generation is not billed; a finished one is, even if you dislike it.

Sources

Related posts

More in Pricing

All Pricing posts

Written by Sume