How many images does $20 buy on Sume? 100 at 4K to 808 on GPT low

The short answer

At Sume's image prices, $20 buys 100 Nano Banana 2.1 images at 4K ($0.20 each), 808 GPT Image 2.5 images at low 1K ($0.02475 each), and 800 images from any of the 2.5-cent rows. The rest fall between. Counts below are whole images, rounded down.

Images per $20

Each count is 20 divided by the price per image, rounded down. Prices are Sume's lead-verified list at default options as of 2026-10-09; the live price for your request comes from GET /v1/images/models/{model_id}/endpoints.

Reading the table

Prices that look close can still differ in whole images: at $0.05 the count is 400, and at $0.055625 it is 359, a 41-image gap on the same $20. The GPT Image 2.5 rows are tiered by quality and size, so low 1K, medium 2K and high 4K are three listed points, not one price. Nano Banana Pro at 4K is $0.375, which gives 53 images.

Failed generations are not charged, so the counts above are for images you receive.

Using the table

Divide your budget by the price of the row you plan to use, then round down. The table does this for $20. For a different budget, scale linearly: $50 buys 2.5 times the count, and $5 buys a quarter.

Mix tiers on purpose. A cheap row for drafts and a high row for the final pick is usually cheaper than rendering everything at the top tier.

Using it

Decide how many finals you need first, then work back. If you need 100 finals at 4K, $20 covers it exactly on Nano Banana 2.1, with no room for retakes. A better plan is to draft at 800 images' worth of cheap rows and re-render the keepers. For example, 200 drafts at $0.025 cost $5.00, leaving $15.00 for 75 images at $0.20.

Do not forget that failed generations are not billed on Sume, so a failed call does not reduce the number of images the $20 buys. The count in the table is a ceiling for successful images at that row, and rounding is down because a fraction of an image cannot be delivered.

Keep a small reserve, such as 5% of the budget, for retries of images that came out wrong. On $20 that is $1, which is 10 more images at the 4K Nano Banana 2.1 price of $0.20? No: $1 buys 5 of those, and 40 at the 2.5-cent rows.

Rounding matters at the edges. For example, $20 divided by $0.20 is exactly 100, but $20 divided by $0.0375 is 533.3, so you get 533 images and $0.0125 is left over. Always floor the result.