Calendar and laptop on a desk marking the October 23 OpenAI model shutdown deadline

OpenAI Shuts Down GPT-4, o1 and o3-mini on October 23

Short answer: On October 23, 2026, OpenAI shuts down gpt-4, gpt-4-turbo, gpt-3.5-turbo, o1, o1-pro, o3-mini and o4-mini in its API. Calls to those names will start returning errors. OpenAI’s deprecations page names a replacement for each one. Search your code for the old names this week. Fine-tuned models on the same bases die the same day.

Four days. After that, a request that says "model": "gpt-4" stops working, and nothing in your logs will warn you until it does.

The shutdown covers a long list of older snapshots, and it was announced back on April 22, 2026. Most teams filed it under “later.” Later is Friday.

Jump to a Section

Which OpenAI models stop working on October 23, 2026?

OpenAI’s deprecations page lists twelve groups of legacy API models with a shutdown date of October 23, 2026, including gpt-4, gpt-4-turbo, gpt-3.5-turbo, o1, o1-pro, o3-mini, o4-mini and gpt-image-1, per OpenAI’s API deprecations page.

When I checked the page line by line, I found that the date also reaches the dated snapshots and the “-completions” variants, not only the short aliases. So gpt-4-0613 and gpt-4 go together, and so do o1-2024-12-17 and o1.

Two things are not on the October 23 list. The newer GPT-5 snapshots retire later, on December 11. And ChatGPT itself is a separate product with its own schedule.

This part trips a lot of people up: the alias is not safe just because you never typed a date.

What does OpenAI say to switch to?

For the October 23, 2026 shutdowns, OpenAI’s deprecations page recommends gpt-5.6-sol for gpt-4, gpt-4-turbo, gpt-4o-2024-05-13, o1 and o3-mini, gpt-5.6-terra for gpt-3.5-turbo and o4-mini, and gpt-5.6-luna for gpt-4.1-nano.

Old model name Shutdown OpenAI’s listed replacement
gpt-3.5-turbo, gpt-3.5-turbo-0125 Oct 23, 2026 gpt-5.6-terra
gpt-4, gpt-4-0613, gpt-4-1106-preview Oct 23, 2026 gpt-5.6-sol
gpt-4-turbo, gpt-4-turbo-2024-04-09 Oct 23, 2026 gpt-5.6-sol
gpt-4.1-nano Oct 23, 2026 gpt-5.6-luna
gpt-4o-2024-05-13 Oct 23, 2026 gpt-5.6-sol
o1, o1-2024-12-17 Oct 23, 2026 gpt-5.6-sol
o1-pro Oct 23, 2026 gpt-5.6-sol with reasoning.mode: pro
o3-mini Oct 23, 2026 gpt-5.6-sol
o4-mini Oct 23, 2026 gpt-5.6-terra
gpt-image-1 Oct 23, 2026 gpt-image-2.5-sunburst or gpt-image-2.5-flare

Source: OpenAI API deprecations page, read October 19, 2026. That page was written in April, so confirm the current model IDs in your own account before you paste them in.

Calendar and laptop on a desk marking the October 23 OpenAI model shutdown deadline
OpenAI retires 12 groups of legacy API models on October 23, 2026.

I noticed that the page’s own replacement names (the 5.6 series) are older than the newest models OpenAI has since shipped. Treat the table as the official minimum, not a recommendation to stop looking.

Your model list endpoint tells you what your account can call today. Use it.

What will my app show after the shutdown?

After October 23, 2026, an OpenAI API request that names a retired model will fail with a client error instead of falling back to a newer one, which means the app that sent it simply breaks.

The exact error text can vary by endpoint and library, so I won’t promise a string here. What you’ll see in practice is a failed call, not a degraded one. A chatbot goes quiet. A nightly summarizer returns empty. A cron job exits non-zero.

If you’ve ever had a pipeline fail at 3 a.m. because of a name in a config file, you know the feeling. This is that, on a schedule you can see coming.

What happens to fine-tuned models?

Fine-tuned models built on a retired base stop working on the same day: OpenAI lists ft-gpt-3.5-turbo, ft-gpt-4, ft-gpt-4.1-nano, ft-babbage-002, ft-davinci-002 and ft-o4-mini for October 23, 2026.

That one is easy to miss, because the model name in your code looks like ft:gpt-4...:your-org::abc123 and no search for “gpt-4” in a model name field will flag it unless you search the prefix.

OpenAI points fine-tuned users at a newer base model. There is no automatic copy of your training. You retrain, or you drop the fine-tune and rely on prompting. Budget time for that, because it is not a one-line change.

How do I find and swap the model names?

To find every retired model name in a codebase, search the repo and your deployed environment variables for each name on the list, then change each one and run a real request against the new name before October 23.

  1. Grep the repo. Search for gpt-4, gpt-3.5, o1, o3-mini, o4-mini and gpt-image-1. Expect false hits on o1.
  2. Check environment variables. A model name often lives in .env, a secrets manager or a Vercel/Cloud dashboard, not in code.
  3. Check no-code tools. Zapier, Make and n8n flows pin a model in a dropdown that nobody revisits.
  4. Pick the replacement. Use the table above, then confirm the ID against your account’s model list.
  5. Test with a real prompt. Run last week’s actual input, not “hello.” Reasoning models behave differently on temperature and token limits.
  6. Deploy before the 23rd. Not on it.

Honestly, the swap itself is usually the easy part. The surprise is prompts that were tuned for GPT-4 phrasing and now return longer or differently formatted answers.

Where the model name hides How to check Risk if missed
Application code Repo-wide search Hard failure on Oct 23
.env and secrets manager List variables, look for MODEL Failure with no code change to blame
Zapier, Make, n8n flows Open each OpenAI step Silent automation stops
Fine-tuned model IDs Search for the ft: prefix Needs retraining, not a rename
Developer reviewing API code on a laptop before swapping model names
Model names usually hide in config files and no-code tools, not only in source code.

What else is on the retirement calendar?

As of October 2026, OpenAI’s deprecations page also lists the Evals platform going read-only on October 31, the Reusable prompts API and Agent Builder shutting down on November 30, and GPT Image mini and 1.5 models retiring on December 1.

  • October 31, 2026: existing Evals become read-only.
  • November 30, 2026: Reusable prompts (v1/prompts), the Evals dashboard and API, and Agent Builder shut down.
  • December 1, 2026: gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest retire.
  • December 11, 2026: the GPT-5 and o3 snapshots retire.

If you’re touching your model names anyway, check these now. Doing it all at once beats four separate fire drills.

Tips

  • Put the model name in one config value so the next retirement is a single edit.
  • Keep a short log of which prompts you retested. It saves an argument later.
  • For bulk, cheap tasks, test the smaller replacement first and compare cost to quality.
  • Read the deprecations page again before you deploy. Dates and recommendations get updated.

Warnings

  • A fine-tune cannot be renamed onto a new base. Plan retraining time.
  • Do not assume an alias will quietly point to something newer. Retired is retired.
  • Replacement models can return longer answers and use different token counts, so budget alerts may trip.
  • Third-party summaries of this deadline disagree on scope. Trust OpenAI’s page over any recap, including this one.

FAQ

Is gpt-4o still available after October 23, 2026?

Only the dated snapshot gpt-4o-2024-05-13 is on the October 23 list. Other gpt-4o entries are not listed for that date, but check the deprecations page for your exact model ID.

Do I need to migrate if I only use ChatGPT, not the API?

No. This shutdown is for the OpenAI API. ChatGPT model availability is handled separately and follows its own announcements.

Will OpenAI redirect old model names to the new ones?

The deprecations page lists a recommended replacement but describes the old names as shut down. Plan on updating the name yourself rather than relying on any redirect.

Sources

📌 Hub guide: For every fix, buying decision, and work-from-anywhere setup in one place — see the Tech & Digital Hub.

Fact-checked against OpenAI’s API deprecations page as of October 19, 2026. Model names, dates and replacements may change after publication.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *