GPT Image 2.5 on fal runs $0.00402 to $0.40026 per image: on Sume
fal's ChatGPT Image 2.5 Flare page lists $0.00402 for 1024x768 low and $0.40026 for 3840x2160 max. At Sume's 1.25 multiplier that is about $0.005 to $0.50.
The fal page for ChatGPT Image 2.5 Flare (read 2026-10-08) gives a cost range of $0.00402 for a 1024 by 768 image at low quality to $0.40026 for 3840 by 2160 at max. Multiplied by the 1.25 that Sume's price book applies to provider list, the same two images come to about $0.0050 and $0.5003. That is a 100-fold spread on one model, and it is why Sume's catalog line for this row is only a display rate.
The spread comes from token billing. The Sume docs give the Fal rates behind both ChatGPT Image 2.5 ids, Flare and Sunburst: $30 per million output image tokens, $8 per million input image tokens and $5 per million input text tokens.
Two ends and the middle
The docs also give the 1024 by 1024 output prices for the two top tiers: $0.09366 at xhigh and $0.21072 at max, before input tokens and Sume pricing. Fal rounds the total up to $0.0001. Multiplying those by 1.25 gives the billed figures in the table, still before input tokens.
| Setting | fal list | x 1.25 (before input tokens) |
|---|---|---|
| 1024 x 768, low | $0.00402 | About $0.0050 |
| 1024 x 1024, xhigh | $0.09366 (output only, Sume docs) | About $0.1171 |
| 1024 x 1024, max | $0.21072 (output only, Sume docs) | About $0.2634 |
| 3840 x 2160, max | $0.40026 | About $0.5003 |
Two traps in the settings
Leaving quality out is not the same as auto. If you omit it Sume sends high; auto reserves the max price for the admission hold, and size auto or a named preset without a verified mapping reserves the upper bound of output tokens. Send an explicit quality and an explicit size when you plan a batch.
Custom sizes have rules: both edges multiples of 16, longest edge at most 3840, aspect ratio at most 3:1, and 655,360 to 8,294,400 pixels. 3840 by 2160 is exactly the pixel cap; 1024 by 768 is just above the floor.
What to do with the range
Treat the low corner as a draft price and the high corner as a deliverable price. Keep the draft at low quality and a small size, and re-run only the keepers at the top. Read usage.cost on every response and total it, because on this row the prompt length and any reference images move the charge, and the endpoint's pricing line is the display rate for one high-quality 1024 image.
For a batch plan, write down three numbers before you start: the draft price, the final price and the share of drafts you expect to keep. Ten keepers out of a hundred at the draft and final prices above is a very different bill from a hundred finals. Then check the first few usage.cost values against your estimate and stop if they are far apart.
Sources
Related posts
More in Pricing
- Does gpt-image-2.5 cost more at 4K? Prices from 1024 to 3840x2160
gpt-image-2.5 price by pixel size on Sume: 1024x1024, 1536x1024, 2048x2048 and 3840x2160 at medium and high, plus the custom-size rules from the docs.
- GPT Image 2.5: leave quality blank and you pay 7 cents, not 1
On Sume, GPT Image 2.5 defaults to high quality when you omit the field. A 1024-class image is 7 cents at high, 2 at medium and 1 at low. 100 images: $7 vs $1.
- GPT Image 2.5 quality auto: 26.35 cents reserved vs 6.59 at high
Leave GPT Image 2.5 on quality auto and Sume estimates the max-quality price: $0.2635 for a square image against $0.0659 at high. Pin quality to control it.
- GPT Image 2.5 for packaging text: default high, xhigh and max prices
GPT Image 2.5 on Sume defaults to quality high. At 1024x1024 the provider output is $0.09366 at xhigh and $0.21072 at max, about $0.117 and $0.263 billed.
Written by Sume