What $12.50 of image credit buys on Sume: 500 down to 33 images

A $12.50 Sume wallet buys 500 Grok Imagine images at $0.025 each, 250 Recraft V4 images at $0.05, or 56 GPT Image 2.5 high 4K images at $0.2225.

5 min readSume
All posts

$12.50 buys 500 Grok Imagine images at $0.025 each, 333 FLUX.2 Pro images at $0.0375, 125 Nano Banana 2.1 images at the 1K tier ($0.10), or 56 GPT Image 2.5 high-quality 4K images at $0.2225. The spread between the cheapest and the most expensive row is 15x, so the model and the tier matter far more than the count you think you need.

Counts below are the whole number of images that fit, and the last column is the balance left over. They use Sume list prices for default options, output only. Image edits with reference images add input tokens on token-priced rows, and Sume's catalog formula rounds billed amounts up to the cent.

Images per $12.50

The price column is Sume list price per image (provider list times 1.25) as of 2026-10-09. The images column is floor(12.50 / price).

What $12.50 buys per image row, as of 2026-10-09
RowPrice per imageImagesLeft over
grok-image$0.0250500$0.0000
imagen-4-fast$0.0250500$0.0000
flux-2-pro$0.03750333$0.0125
seedream-5-lite$0.04375285$0.03125
recraft-v4$0.0500250$0.0000
ideogram-v4.5 (medium)$0.0750166$0.0500
nano-banana-2.1, 1K$0.1000125$0.0000
nano-banana-pro, 1K$0.1875066$0.1250
gpt-image-2.5, high 4K$0.2225056$0.0400
nano-banana-pro, 4K$0.375033$0.1250

Reading the table

Two rows tie at $0.025: Grok Imagine and Imagen 4 Fast. They are different tools. Imagen 4 Fast is text-to-image only in Sume's catalog, so it cannot take a reference photo. Seedream 5 Lite at $0.04375 sits between FLUX.2 Pro and Recraft V4. Nano Banana Pro is flat at $0.1875 for 1K and 2K and doubles to $0.375 at 4K, so a 2K image costs the same as 1K.

GPT Image 2.5 is the odd one because its price depends on both quality and size. The high 4K figure used here is $0.2225. A low-quality 1K image is $0.02475, which would make the same $12.50 buy 505 images.

  • Test prompts: spend the first $1 on a $0.025 row and keep the survivors
  • Finals: spend on one of the $0.10 to $0.2225 rows only for the prompts that passed
  • Unknown model: ask GET /v1/images/models for supported_parameters before you pin a ratio or tier

Turn it into a rule

A simple budget rule is to split a wallet 20 / 80 between exploration and finals. Of $12.50, that is $2.50 of drafts (100 images at $0.025) and $10.00 of finals (100 images at $0.10, or 44 at $0.2225). Adjust the split to your pass rate: if only one in four drafts survives, a 100-draft round supports about 25 finals.

Check the balance before a batch

Sume reserves the estimated cost of a generation when you submit it. The generation admission docs list 402 insufficient_credits as the response when Sume cannot reserve the estimated cost from the workspace balance, before any provider work starts. With a $12.50 wallet, a batch of 60 GPT Image 2.5 high 4K images needs 60 x $0.2225 = $13.35, so the last few submissions would be refused.

Two habits avoid that surprise. First, keep a small margin because Sume's catalog formula rounds billed amounts up to the cent. Second, if you use auto quality or an auto size on GPT Image 2.5, the docs say the reservation uses the upper bound, so set quality and size yourself to reserve only what you intend to spend.

Sources

Related posts

More in Pricing

All Pricing posts

Written by Sume