GPT Image 2.5 4K image price: 4 cents medium, 13 cents high on Sume

A 3840x2160 GPT Image 2.5 image costs 2 cents at low quality, 4 at medium and 13 at high on Sume. At high quality that is about 2.6 times a 1920x1080 image, which costs 5 cents.

Price by size

GPT Image 2.5 is priced from estimated output tokens ($30 per 1M image tokens), so size changes the bill. The table prices explicit pixel sizes.

A surprise in the table

The 1024x1024 square costs more at high quality (7 cents) than 1920x1080 (5 cents). The estimator maps each size to a token count, and the wide frame maps to fewer tokens than the square. Do not assume more pixels always means a higher price; run the estimate for your exact size.

Auto sizes and presets without a verified token mapping reserve the upper bound, according to the Sume image models docs, so pass explicit width and height when you can.

When to use 4K

Use 4K when the output is printed or cropped hard. For screens, 1920x1080 at high is 5 cents against 13. For the batch math see the four-image price post.

• Name a size and a quality.
• Check the estimate for odd sizes.
• Prices from the public estimator.

How to choose a size

Start from the output, not the model. An image that ends up in a 1920x1080 slide gains nothing from 4K, and an image printed at poster size needs it. The price difference at medium is 2 cents against 4, so the real cost of picking 4K by habit is small per image but adds up: 10,000 images at medium is $200 at 1080p and $400 at 4K.

Also consider the cap on pixel counts. The image schema accepts specific presets and the auto sizes, so a custom size can be rejected if it falls outside what the model supports. Test one request at your exact dimensions, read the estimate, and then scale. The image models docs list the options.

How these numbers were produced

The figures are Sume's billed-amount estimates for gpt-image-2.5 on 2026-10-06, at 1024x1024, 1920x1080 and 3840x2160 at low, medium and high quality. The tables show whole cents.

These are estimates of the billed amount for the stated settings. Anything not in the table, such as a different frame shape, a reference input or a changed quality tier, can change the number, so run the estimate for your own request before a large batch. Where this page and the public API pricing list disagree, the pricing page and the live quote win.