GPT Image 2.5 cost by aspect ratio: square, 3:2, 4:5, 16:9

On GPT Image 2.5 the shape you ask for changes the output-token estimate. At medium, 1024x1024 is $0.01317, 1536x1024 (3:2) is $0.01029, 1024x1280 (4:5) is $0.01134 and 1536x864 (16:9) is $0.0084, before input tokens and Sume pricing. A square is the dearest of these because it has the most pixels on the short edge. Choose the shape the placement needs, not the shape that looks cheapest.

Estimates by shape

The numbers use the output-token estimate from OpenAI's size and quality calculator, which the Image API docs say Sume's estimates follow, at $30 per million output tokens, the rate listed on Fal's GPT Image 2.5 Flare page, and sizes that satisfy the rules in OpenAI's image generation guide. The 3:2 column is 1536x1024, which OpenAI lists as a recommended landscape size.

What to take from it

Against the square at the same tier, 3:2 is about 22% cheaper, 4:5 about 14% cheaper and 16:9 about 36% cheaper. The saving is fractions of a cent at low and medium and noticeable at max.

The estimate depends on the short edge as well as the total pixels, which is why a frame with a shorter short edge costs less even when its pixel count is close to the square's.

Pick by placement

• Feed square or carousel: 1:1.
• Mobile portrait: 4:5; ask for 1088x1360 if you need exact pixels, see the 4:5 post.
• Photo-like landscape: 3:2.
• Video thumbnails and headers: 16:9.
• Edits: aspect_ratio: "auto" keeps the reference's shape, so the shape is whatever you sent.

Send the ratio

aspect_ratio takes the normalised ratios in the docs; read the model's descriptors first, because a model only accepts the values it lists.

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": "A flat-lay of coffee beans and a brass scoop",
    "aspect_ratio": "3:2",
    "quality": "medium"
  }'

Caveat

These are estimates for output tokens. Verify with usage.cost on a real call; Sume applies its own pricing, and the endpoint pricing lines are what your wallet is charged.