GPT Image 2.5 at 1024: xhigh is about 3,122 tokens and max about 7,024

The Sume Image API docs say GPT Image 2.5 output is priced at $30 per million output image tokens and that, at 1024 x 1024, xhigh output is $0.09366 and max output is $0.21072 before input tokens and Sume pricing. Dividing by the token rate gives 0.09366 / 0.00003 = 3,122 tokens for xhigh and 0.21072 / 0.00003 = 7,024 tokens for max. These counts are derived from the documented prices, not stated by OpenAI or Sume.

With Sume's 1.25 multiplier, the same images are 0.09366 x 1.25 = $0.117075 and 0.21072 x 1.25 = $0.2634, and the catalog formula rounds billed amounts up to the cent: $0.12 and $0.27. A max-quality 1024 image therefore costs more than a high-quality 4K image on the Sume list ($0.2225).

Implied tokens for the other tiers

The same arithmetic applied to the three list prices that the Sume price list uses for GPT Image 2.5 gives the table below. Each implied figure is the Sume price divided by 1.25, then divided by $30 per million. It assumes the price is all output tokens, which is true for a prompt-only request and understates tokens when references add input.

What the numbers say about quality

Quality is a token budget. Going from xhigh to max at 1024 roughly doubles the tokens (7,024 / 3,122 = 2.25), and the price follows. If your prompt is a poster with dense text, spend the extra on one proof. If it is a simple product on white, a lower tier may look identical for a fraction of the cost.

The same page of docs says auto quality reserves max, and that auto size or a named preset without a verified pixel mapping reserves the upper bound of output tokens. A reservation is a hold on balance, not necessarily the final charge, but a small wallet can fail with 402 insufficient_credits on a request that would have been cheap. Set quality and size yourself.

• xhigh at 1024: about $0.117 before cent rounding
• max at 1024: about $0.263 before cent rounding
• Auto quality: reserves the max tier, so pin quality in code

Check the real bill

Run one image per tier you plan to use and read usage.cost. The docs state that Sume meters image models per image and that token counts in usage are 0 in v1, so the tokens in this post are a way to understand the price, not a field you can read back.

Worked example: 60 images

Suppose a campaign needs 60 images at 1024 x 1024 and quality is the open question. At xhigh, 60 x $0.117075 = $7.0245 before cent rounding. At max, 60 x $0.2634 = $15.804. The difference is $8.78, which is the price of asking for 3,902 more tokens on each image (7,024 minus 3,122). Run five of each, compare them side by side, and only then decide which tier the other fifty-five need.