Three OpenAI image ids end Dec 1: replacements and 1,000-image cost

gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest shut down 2026-12-01. OpenAI names gpt-image-2.5-sunburst or -flare; Sume's rates for 1,000 images.

5 min readSume
All posts

OpenAI's deprecations page lists three image ids shutting down on 2026-12-01: gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest. The replacements it names are gpt-image-2.5-sunburst or gpt-image-2.5-flare. Sume lists both as openai/gpt-image-2.5-sunburst and openai/gpt-image-2.5 (Flare), so the swap is a model-id change in the request.

What to change

OpenAI's page announces the change on 2026-06-02 and gives one replacement pair for all three ids; it does not map one old id to one new id. That means you pick the variant, not the page. On Sume the two variants use the same token rates and the same options: up to 16 reference images, an optional mask_url, and quality of auto, low, medium, high, xhigh or max. If you omit quality, Sume defaults to high.

  • gpt-image-1-mini to openai/gpt-image-2.5 or -sunburst, then set quality to low or medium if you used mini for cost.
  • gpt-image-1.5 to the same pair; keep high for parity with the old default.
  • chatgpt-image-latest is an alias that moves; pin an explicit id instead.
  • Check every hard-coded id before 2026-12-01, not just the ones in your main service.

The 1,000-image arithmetic

Sume bills provider list times 1.25. The three rows below are the catalog's low, medium and high examples at the sizes shown; other sizes and quality values price differently, and the catalog's default estimate is the high row.

Sume gpt-image-2.5 per-image rates and a 1,000-image run (catalog read 2026-10-09)
QualitySizePer imagex 1,000 images
low1K$0.02475$24.75
medium2K$0.055625$55.625
high4K$0.2225$222.50

What stays the same and what does not

The replacement is a different model generation, not a renamed one, so outputs will not match the old ids pixel for pixel. Run your prompt set through both before the cutover and compare. Sume reserves the price at submit and reports the billable amount on the job, so a test batch tells you the real cost of your own sizes. For the alias-pinning side see pinning a model id.

A checklist before December 1

Search your code, config and prompt templates for the three ids. Search environment files and feature flags too, since an id often lives outside the code. Replace each with an explicit new id, run your regression set at the quality level you actually ship, and compare cost at that level. If you used chatgpt-image-latest because you wanted whatever was newest, decide now whether you prefer pinned behavior, because the alias is going away either way.

OpenAI's page announces 2026-06-02 and shutdown 2026-12-01, so the window from announcement to shutdown is 182 days; as of 2026-10-09, 53 days remain.

What I could not confirm

OpenAI's page lists the two replacement ids but not their prices. The Sume rates quoted are Sume's own catalog numbers, not OpenAI's. All vendor figures were read on 2026-10-09 and are listed in the sources; Sume figures come from its public catalog and docs the same day.

Sources

Related posts

More in Comparisons

All Comparisons posts

Written by Sume