Does GPT Image 1 still work on October 7, 2026? 16 days left

Yes, for now. OpenAI's deprecations page, read 2026-10-07, lists the GPT Image 1 shutdown for 2026-10-23, which is 16 days after today, so calls to the model id still need to be migrated before that date. Sume's image catalog does not list gpt-image-1 as a row; the OpenAI image rows it serves are openai/gpt-image-2.5, openai/gpt-image-2.5-sunburst and openai/gpt-image-2.

If your code calls OpenAI directly, the date on OpenAI's page is the one that matters, and you should read it yourself. If your code calls Sume, the question is which of the three ids you should change to.

The dates and the options

The table separates what OpenAI's page says from what Sume lists. OpenAI facts are from the deprecations page; Sume facts are from its image catalog and price tables.

A note on wording. Some tools and wrappers keep a cached model list, so a model can look available in a dropdown long after the provider has scheduled it for removal. Treat the provider's page as the source for the date and your own request logs as the source for what you actually call. If a log line shows gpt-image-1 in the last 30 days, that call site is live and needs a change.

Also look at indirect callers. A third-party plugin, a no-code automation or a scheduled export may send the old id without appearing in your repository. Ask each team which tools generate images for them, and check each tool's settings for a model field. These are the cases that fail on the shutdown date because nobody owned them.

For image jobs that run rarely, such as a quarterly report or a yearly campaign, set a calendar reminder to run the test call now, not when the job fires. A test call costs a fraction of a cent at the low tier and tells you whether the new id, response handler and size all work.

What to do this week

Day one: search your repo and your stored configs for gpt-image-1, including gpt-image-1-mini. Day two: pick the replacement id, set quality explicitly and run a small batch. Day three onward: ship the change behind a flag, watch usage.cost, and retire the old id. A stored post on this site has a Python scan for the search step.

Write down the owners. Each call site belongs to a team, and the team that owns the code that sends the request is the team that has to test the response. A shared image helper is the easiest case: change it once, run its tests, and every caller moves. Scattered direct calls are the hard case, which is why the search comes first.

Do not wait for the last day. If OpenAI's page changes its schedule, you will want the change already tested, and a shutdown during a campaign is a bad time to learn that your size field was wrong.

Which id on Sume

For most jobs, openai/gpt-image-2.5 is the default, with the lowest per-image prices. Use openai/gpt-image-2.5-sunburst if you want the second variant, which carries the same price and limits in Sume's catalog. Use openai/gpt-image-2 if you have approved its output and want to keep it, at a higher price. The request below is the shape to test.

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": "Studio photo of a leather notebook on a desk",
    "aspect_ratio": "4:3",
    "quality": "medium"
  }'

A 16-day plan

At the cheapest tier, a full regression of 200 prompts on openai/gpt-image-2.5 costs $1.48 billed, and at medium $3.30. That is small next to the engineer time spent on the migration, so test widely and cheaply.

Allocate four days to find and list every call site, four to run the replacement on real prompts, four to roll out and four of slack. The rollout is a model id and a response handler change. The slack is for the unexpected: a hard-coded size, a base64 decoder, or a job that only runs monthly.