Re-render a gpt-image-1 library before Oct 23: cost on 2.5

Re-rendering 1,000 images at 1024x1024 on GPT Image 2.5 costs about $7.35 at low quality, $16.46 at medium and $65.85 at high on Sume, and the default quality, which is high, is the one you pay if you leave the field out. OpenAI's deprecations page, read on 2026-10-05, lists the gpt-image-1 shutdown for October 23, 2026 and names gpt-image-2.5-sunburst or gpt-image-2.5-flare as the replacement. That leaves 18 days.

What the deprecation page says

The OpenAI deprecations page lists the legacy GPT model snapshots, announced April 22, 2026, with gpt-image-1 shutting down on October 23, 2026 and a recommended replacement of gpt-image-2.5-sunburst or gpt-image-2.5-flare. This post does not compare against OpenAI's own prices, because that page was not read for them; it prices the replacement on Sume.

The per-image price by quality

Sume's image docs say Flare and Sunburst use the fal token rates: $30 per million output image tokens, $8 per million input image tokens and $5 per million input text tokens. At 1024x1024 the fal list prices are $0.00588 for low, $0.01317 for medium, $0.05268 for high, $0.09366 for xhigh and $0.21072 for max, before input tokens. Sume bills list times 1.25. The columns show a single image, then 1,000 and 10,000 images, for a prompt with no input images.

Two defaults that change the bill

If you omit quality, the Sume default is high, which is $0.06585 per image at 1024x1024. If you send quality: auto, Sume reserves the max price of $0.2634 per image at submit, so the balance must cover that even if the job settles lower. A migrated script that used a single default on the old model can therefore cost four times as much on 2.5 at the default as at medium, unless you set the quality explicitly.

• Pin quality to the level you actually need.
• Avoid auto in batch jobs, because it reserves max.
• Estimates exclude input tokens, so edits with reference images cost a little more.

A migration plan with a budget

Take a library of 2,000 images. Re-rendering everything at medium is about $32.93. Re-rendering the 10 percent that you still use in production at high is $13.17, plus the other 90 percent at low for $13.23, for about $26.40 in total. Setting quality per image costs a few lines of code and saves more than the cost of the slowest hour of the migration.

Which of the two replacement ids you pick does not change the price: the docs state that Flare and Sunburst use the same fal token rates, and that automatic model routing uses Flare. Pick one, pin it, and keep it constant across the library so the images match in style.

Eighteen days is enough for a batch run, and not enough to discover in the last week that your script reserved max on every call. Test with ten images first: ten at medium cost about $0.16, and ten at max would cost $2.63, which is a cheap way to see the spread before you commit a thousand. Log the quality you sent next to every output so you can explain a surprising bill later.

Before the cutoff

Image size matters as well as quality. These figures are for 1024x1024; larger outputs use more tokens, and Sume's estimate for named presets without a verified pixel mapping reserves the upper bound. Keep to the size you need. Check the date and the replacement ids on OpenAI's page, then the live price on Sume's pricing. The fal token rates are on fal's page and are what the Sume figures build on.