GPT Image 2.5 API price: token rates and per-image estimates

GPT Image 2.5 is priced in tokens: $30 per million output image tokens. See the per-image reserve on Sume at 1024, 1920x1088 and 4K for each quality level.

5 min readSume
All posts

The GPT Image 2.5 API is priced by tokens, not by image: OpenAI lists $5 per million text input tokens, $8 per million image input tokens and $30 per million image output tokens for both Flare and Sunburst. What one picture costs depends on its size and its quality setting, because those decide how many output tokens it uses.

On Sume, both ids (openai/gpt-image-2.5 and openai/gpt-image-2.5-sunburst) use the same rates. The table below is the amount Sume reserves on submit for one image, output tokens only, computed from Sume's pricing code. Read on 2026-09-29; the Image API docs explain the rest.

What does one GPT Image 2.5 image cost on Sume?

It ranges from about a cent at low quality to about half a dollar for a 4K image at max. Each figure is the reserve for one output image, rounded up to a whole cent, with Sume's margin applied. It leaves out input tokens for your prompt and reference images, which are added on top.

Reserve estimate per output image on Sume, from pricing code, read 2026-09-29.
Quality1024x10241920x10883840x2160
low$0.01$0.01$0.02
medium$0.02$0.02$0.04
high$0.07$0.05$0.13
xhigh$0.12$0.09$0.23
max$0.27$0.20$0.51

Why does a 1920-wide image price lower than 1024x1024 at high quality?

Because the output token count follows OpenAI's size and quality calculator, not the pixel count alone. fal's pricing page for Flare shows the same pattern in its own high-quality examples: $0.03960 at 1920x1080 against $0.05268 at 1024x1024 (provider prices, before Sume's margin). Do not assume a bigger picture costs more; read the row for the size you will send.

What happens if I leave quality or size on auto?

Sume reserves the worst case. quality: "auto" reserves as max, and auto size, or a named preset with no verified pixel mapping, reserves the output-token upper bound. Omit quality and the request runs at high. If you know the tier and size you need, set both, so the reserve matches what you expect to pay.

Where do I see the price before I call?

GET /v1/images/models returns each model's pricing line. Real requests are estimated from their own size and quality, so use the table above for a specific size. The response of a finished call carries usage.cost, the USD amount billed to your wallet. A failed or cancelled generation is not billed.

curl "https://api.sume.com/v1/images/models" \
  -H "Authorization: Bearer $SUME_API_KEY"

Sources

Related posts

More in Pricing

All Pricing posts

Written by Sume