GPT Image 2 medium costs 4x GPT Image 2.5 medium; low is the same

On Sume, GPT Image 2 costs $0.0664 at medium and $0.2639 at high for a 1024x1024 image, which is about four times the $0.0165 and $0.0659 that GPT Image 2.5 costs at the same tiers. At low the two are nearly equal: $0.0076 against $0.0074, a 3% difference. If you are on GPT Image 2 and not using low, moving to 2.5 is the biggest price change in this lineup.

Prices here are Sume's, which are the provider list rates times 1.25, with the provider total rounded up to $0.0001. Both ids are in Sume's catalog today: openai/gpt-image-2 and the two 2.5 ids, openai/gpt-image-2.5 (Flare) and openai/gpt-image-2.5-sunburst.

Tier by tier

The GPT Image 2 column is Sume's quote for that tier. GPT Image 2 quotes the same figure at high, xhigh, max and auto, since the estimator uses one top-tier price for all four.

The same pattern at other sizes

The gap is not specific to the square. At 1536x1024, GPT Image 2 medium quotes $0.0526 against $0.0129 for 2.5, and at 3840x2160 it quotes $0.1264 against $0.0325. At low and 3840x2160 the two are $0.0151 and $0.0140, again within 10%.

Do the sums on your own volume before you decide. At 10,000 images a month, medium at 1024x1024 is $663.75 on GPT Image 2 and $165.00 on 2.5, a difference of $498.75 a month. The reason is how each model is priced. Sume's Image API page says GPT Image 2.5 uses Fal token rates and the OpenAI size and quality calculator, which is why its price climbs steeply with quality. GPT Image 2 keeps its older table, where the middle and high tiers sit much higher.

What this means for a migration

If you came to gpt-image-2 from gpt-image-1, you have more than one move available, and the price answers which to pick.

• From gpt-image-1: OpenAI's deprecations page (read 2026-10-05) names the 2.5 ids as the replacement, and gpt-image-1 shuts down on 2026-10-23. Going to 2.5 is the supported path and the cheaper one.
• From GPT Image 2 at medium or high: moving to 2.5 at the same tier cuts the price by about 75%. Test the output first, since the price drop is not a promise that the image is the same.
• From GPT Image 2 at low: the price barely changes, so move only if you want the 2.5 features such as xhigh, max and background.

Checks before you switch

Run your regression prompts at the same tier on both models and keep the outputs side by side. A tier is a name, not a guarantee, so a 2.5 image at medium is not a pixel-for-pixel copy of a GPT Image 2 image at medium, and any downstream step that depends on a particular look should be tested again.

Check the size rules too. GPT Image 2.5 accepts custom sizes with both edges as multiples of 16, a longest edge up to 3840 and a 3:1 ratio limit. If you send a size that was valid for GPT Image 2 and fails these rules, Sume returns an error instead of resizing the request.

Finally, check any code that reads the price. A budget guard set around a GPT Image 2 medium image of about six and a half cents will never fire on a 2.5 image of one and a half cents, so lower the guard to match the new figure.

Request

Switching is one field: change the model value from openai/gpt-image-2 to openai/gpt-image-2.5. Keep quality and image_size explicit so the quote matches the table, and compare usage.cost in the first response with the figure above. A request that carries a parameter the new model does not accept returns 400 unsupported_parameter, so a leftover field shows up in the first test and not in production.

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": "Flat lay of three notebooks on a grey desk",
    "quality": "medium",
    "image_size": "1024x1024"
  }'