GPT Image 2.5 at 2560x1440 costs 5% more than 1024x1024 at high

At high quality, a 2560x1440 GPT Image 2.5 image quotes $0.0691 on Sume, 4.9% more than the $0.0659 for 1024x1024, while carrying 3.5 times the pixels. At xhigh and max the difference is also about 5%. Only medium is a bigger step, at 9%.

The reason is how the output tokens are counted. Sume's Image API page says output estimates use OpenAI's size and quality calculator at $30 per million output image tokens, and that calculator grows far more slowly with pixels than with the quality tier. Quality is the cost lever; resolution mostly is not. That is the opposite of how most people budget, since they assume a larger image costs proportionally more. The table below puts the two sizes next to each other so you can check the claim for your own tier before you change a default.

Square against 2560x1440, by tier

Each row compares the same quality at the two sizes. All amounts are Sume's quote with a one-character prompt.

What this means for your defaults

If you were holding output at 1024 to save money and stay at high, the saving is about 5%, which is a small price for an image with more than three times the pixels. Check the larger render first when the output is going to be cropped, zoomed, or printed, since the extra detail is nearly free. Do the same test with a second prompt that contains small text, because fine detail is where extra pixels help most and where a reviewer will notice a difference first.

The saving from lowering quality is far larger. Dropping from high to medium at 1024x1024 takes the quote from $0.0659 to $0.0165, a cut of 75%.

The limits that still apply

The size has to be valid before the price matters.

• Both edges must be multiples of 16: 2560 and 1440 are.
• The longest edge is at most 3840, the aspect ratio at most 3:1, and the pixel total between 655,360 and 8,294,400. 2560x1440 is 3,686,400 pixels.
• A named preset without a verified pixel mapping reserves the upper bound of output tokens, so pass image_size: "2560x1440" as pixels when you want the figure in the table.

Request

The call below prices at the 2560x1440 row. Check usage.cost in the response against the table.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "Wide studio photo of ceramic bowls on a shelf",
    "quality": "high",
    "image_size": "2560x1440"
  }'

Where the pattern breaks

The pattern is not linear to the top. At 3840x2160 and high, the quote is $0.1251, which is 1.9 times the square price, so very large canvases do cost more. Between 1024 and 2560 on the long side, the penalty is small.

Portrait and landscape images at 1536x1024 are cheaper than the square: at high the quote is $0.0515, which is 22% below the square. A 3:2 frame has fewer pixels than the square at that size, so check the pixel count before you assume a wider frame costs more.

A rule for choosing the size

Pick the size from the use, then the tier from the budget. If the image ships at 1280 pixels on the long side, 2560x1440 is a render you will downsample, and downsampling is a fair way to get a clean result for about five percent more at high.

If the image is a background or a thumbnail, 1024 is enough and the premium buys nothing. If it needs to survive a crop, take the larger size. In every case, send the size as pixels so the quote does not fall back to the reserved upper bound.