Does an image cost more at 16:9 than 1:1 on Sume? Flat rows vs GPT 2.5

On most Sume image models, changing aspect_ratio does not change the price: Seedream, Flux, Qwen, Imagen, Grok and Recraft each have one per-image rate in the repo's rate table, and Nano Banana and Ideogram 4.5 price by tier or quality, not by ratio. ChatGPT Image 2.5 is the exception, because it is billed from output tokens that depend on pixels and quality.

Flat per-image rows

ChatGPT Image 2.5 follows pixels

For ChatGPT Image 2.5, Sume estimates the output with OpenAI's size and quality calculator and charges fal's $30 per million output tokens as list, then times 1.25. The same quality costs less on a smaller canvas. Figures below are for high quality, output tokens only.

What to take from it

• A 16:9 thumbnail on GPT 2.5 at 1536x864 is cheaper than a 1024 square at the same quality, because it has fewer pixels' worth of tokens in this estimate.
• On the flat rows, do not squeeze a banner into a square to save money. The price is the same.
• On GPT 2.5, auto quality reserves the max price before the job runs, so set quality yourself when you budget.
• Input images on edits add input tokens on top of the output estimate.

Budget formula

For flat rows, budget is images times the billed rate. For GPT 2.5, estimate per canvas and quality. Forty squares at 1024 and high quality cost 40 x $0.065875 = $2.635, and at low quality 40 x $0.007375 = $0.295. Edits that send reference images add input tokens.

Do not reuse a high-quality 1024 figure for other sizes or qualities. If a spend cap matters, set quality explicitly and check usage.cost on the first call.

Check your own call

Read usage.cost from the response, or the pricing line from GET /v1/images/models/{id}/endpoints. The catalog list for GPT rows is the high-quality 1024 square; other sizes come from the estimator at submit time. The Image API page describes the token rates, and OpenAI's image generation guide lists the custom-size rules that GPT sizes follow.