GPT Image 2.5 cost per megapixel: $0.050, $0.019, $0.012

Per megapixel, GPT Image 2.5 Flare at high quality gets much cheaper as the frame grows: about $0.0502 at 1024x1024, $0.0191 at 1920x1080 and $0.0121 at 3840x2160. The figures come from the three prices on the fal model page. If you need a large image, one big render beats tiling several small ones on price.

The arithmetic

Divide each fal price by the pixel count in millions. The prices are fal list prices, before any Sume margin.

Why it matters for a print or poster job

A 3840x2160 render is 7.9 times the pixels of a 1024x1024 render for 1.9 times the price. The pixel count is also exactly the 8,294,400 maximum that both OpenAI and the Sume docs give for custom sizes. Both edges must be multiples of 16, the longest edge cannot pass 3840, and the aspect ratio cannot exceed 3:1.

This is the cost of pixels only. It does not make an upscale of a small image equal to a native render, because the model draws the extra detail itself.

What this does not cover

These are text-to-image prices. Edits with reference photos add input image tokens at $8 per million, per the Sume docs. Sume bills per request and shows the amount in usage.cost, so use this table to plan and the response to confirm.

• Plan with the per-MP column, then verify with one real job.
• Keep quality at high only for finals; low and medium are cheaper.
• Do not compare against flat-priced rows such as Qwen Image without checking the size you need.