164 days from DALL-E 3 shutdown to GPT Image 1, then 39 more

OpenAI's deprecations page lists the DALL-E removals on May 12, 2026, the GPT Image 1 removal on October 23, 2026, and three more image ids on December 1, 2026. That is 164 days from the DALL-E shutdown to the GPT Image 1 shutdown, then 39 days more to December 1. From today, 2026-10-09, GPT Image 1 has 14 days left and the December 1 group has 53.

The pattern is the useful part. In the May notice, OpenAI named gpt-image-2, gpt-image-1 and gpt-image-1-mini as the replacements for DALL-E. Anyone who picked gpt-image-1 then is on a model that retires within six months. The current page names gpt-image-2.5-sunburst or gpt-image-2.5-flare for every row in the newer notices.

Every image id on the page, by date

The table is from OpenAI's deprecations page, read 2026-10-09. Replacement ids are as OpenAI lists them.

What this means on Sume

Sume's Image API docs list ChatGPT Image 2.5 as openai/gpt-image-2.5 (Flare) and openai/gpt-image-2.5-sunburst, and the Image Router catalog also lists gpt-image-2. None of the retiring ids appears in the catalog's model list as of 2026-10-09. A request with an id that Sume does not list gets a model error, so the change on your side is to the model string, not to the endpoint.

Pin an exact id, and do not rely on an alias like chatgpt-image-latest. An alias that moves lets a provider change your output and your price between two deploys, and OpenAI's own table shows that this alias has its own shutdown date.

A date check you can run

The script prints the days left for each id on the page, so you can drop it in a weekly job. It uses only the standard library.

Move the work in two steps: first the single id that goes first (October 23), then the December 1 group, so a problem in a new model does not land on the deadline day. Keep the old and new ids behind one config value so that a rollback is a one-line change until the old id is actually removed.

Run a small fixed set of prompts through the new id before the cutover and compare outputs by eye, since a replacement model will not draw the same picture from the same words. Record the billed amount for each, because the price per image can also differ between the old and new rows.

from datetime import date

SHUTDOWNS = {
    "gpt-image-1": date(2026, 10, 23),
    "gpt-image-1-mini": date(2026, 12, 1),
    "gpt-image-1.5": date(2026, 12, 1),
    "chatgpt-image-latest": date(2026, 12, 1),
}

today = date.today()
for model, day in sorted(SHUTDOWNS.items(), key=lambda kv: kv[1]):
    print(f"{model}: {(day - today).days} days left")

Price check before the swap

GPT Image 2.5 on Sume lists at $0.02475 for low quality at 1K, $0.055625 for medium at 2K, and $0.2225 for high at 4K. Omit quality and the default is high. Send the quality you want, then read usage.cost on the first few responses. The Image API docs describe the fields.