OpenAI retires GPT-5.5: check your saved ChatGPT and Codex settings

By Jason Sibley

What happened

On 14 October 2026, OpenAI retired GPT-5.5 from ChatGPT, ChatGPT Work and Codex. The OpenAI Help Center Model Release Notes say:

"We're retiring GPT-5.5 from ChatGPT, ChatGPT Work, and Codex. This change does not affect API access."

The ChatGPT Learn model docs spell out the Codex timing. It took effect on 14 October 2026 at noon PDT (19:00 UTC) — 20:00 in the UK and Ireland. It applies to people signed in to Codex with ChatGPT. The OpenAI API is untouched. So are Codex sessions that use an API key.

For Codex with a ChatGPT sign-in, OpenAI names these replacements:

• Plus, Pro, Business, Enterprise and Edu: GPT-6.1 Sol (gpt-6.1-sol), when available.
• Free and Go: GPT-6 Luna (gpt-6-luna) in the desktop app, when available.

It also tells you to replace gpt-5.5 in workspace defaults, saved model settings, managed configurations, custom agents, scheduled tasks, and scripts that pick a model.

Secondary reporting matches that note for consumer, Business, Enterprise and Edu plans across ChatGPT, ChatGPT Work and Codex. Trust the two OpenAI pages for your own plan.

This is not another "new model" post. It is about a model vanishing from the picker, and about anything you saved that still points at it.

What it means for a small business

If you only open ChatGPT for the odd email or proposal tidy-up, you will probably just see a different model in the picker. Choose a current one and move on.

The snag is the stuff you set up once and forgot. Plenty of small teams have a custom agent that drafts quote follow-ups, a scheduled task that summarises the week's enquiries, or a script a freelancer left behind that hard-codes a model name. Anything that still says gpt-5.5 is on OpenAI's fix list.

OpenAI does not say how each saved setting behaves once its model is gone. Do not assume a loud failure, or a clean hand-off. Test it. A job that quietly stops, or quietly runs on another model, is easier to miss than one that errors.

A different model may also write in a different voice. That is my reading, not an OpenAI claim. If your prompts are tuned for a customer-facing tone, read the first few outputs before they go out under your name.

One relief: tools on the API, or Codex with an API key, sit outside this retirement. If an agency or developer built your setup, ask which route it uses. You may have nothing to change.

What you can do this week

1. List where you use ChatGPT, ChatGPT Work or Codex. Note who set each one up. Include the freelancer's script, the shared workspace and that scheduled task from the spring.
2. Search for gpt-5.5. Workspace defaults, saved model settings, managed configs, custom agents, scheduled tasks, scripts. Those are the places OpenAI names.
3. Swap in a current model. For Codex with ChatGPT sign-in, OpenAI points paid plans at GPT-6.1 Sol and Free or Go users at GPT-6 Luna in the desktop app, when each is available. Check your own picker before you change anything.
4. Run each affected job once and read the output. Confirm scheduled tasks still fire, agents still do the job, and customer-facing copy still sounds like you. Note what you changed and when.
5. Confirm the API route. If a tool uses an API key rather than a ChatGPT sign-in, this retirement does not touch it. Ask whoever built it, and write the answer down.

How we can help

If hunting through settings is not how you want to spend an afternoon, we can help. Soft ask only.

• Free website help: if a tool feeds your site, forms or email, we can look and tell you what we see.
• AI sales and marketing desk: bring a workflow that has stopped behaving and work out what needs changing.
• AI sessions for small businesses: plain-English sessions on AI for marketing, sales, email and day-to-day running.

Get in touch via My AI Apprentice if that would help. If the checklist above sorted it, even better.