OpenAI is shutting down gpt-4, gpt-3.5-turbo, o1 and 30+ legacy models this fall
Two shutdown waves are dated on OpenAI's official deprecations page: 4 models on September 28, 2026 and 29 more on October 23, 2026. After the date, API calls to these model ids return errors — and every n8n, Make or Zapier workflow that names one of them stops producing output.
Why automation workflows are the most exposed
Automation platforms make this failure mode sneaky: the model id is typed into a node or module once — often years ago — and nobody looks at it again. A client workflow built in 2024 with gpt-4 or gpt-3.5-turbo hard-typed into the OpenAI node keeps working right up to the shutdown date, then fails on the next run. If the workflow has no error branch, the failure is silent: enrichments stop, replies stop, and the client notices before you do.
Wave 1 — September 28, 2026 (4 models)
| Model being shut down | OpenAI's recommended replacement |
|---|---|
gpt-3.5-turbo-instruct | gpt-5.6-terra |
gpt-3.5-turbo-1106 | gpt-5.6-terra |
babbage-002 | gpt-5.6-terra |
davinci-002 | gpt-5.6-terra |
Wave 2 — October 23, 2026 (29 models, the big one)
| Model being shut down | OpenAI's recommended replacement |
|---|---|
gpt-3.5-turbo · gpt-3.5-turbo-0125 · gpt-3.5-turbo-completions | gpt-5.6-terra |
gpt-4 · gpt-4-0613 · gpt-4-completions · gpt-4-0613-completions | gpt-5.6-sol |
gpt-4-turbo · gpt-4-turbo-2024-04-09 · gpt-4-1106-preview · gpt-4-turbo-completions | gpt-5.6-sol |
gpt-4o-2024-05-13 | gpt-5.6-sol |
gpt-4.1-nano · gpt-4.1-nano-2025-04-14 | gpt-5.6-luna |
o1 · o1-2024-12-17 | gpt-5.6-sol |
o1-pro · o1-pro-2025-03-19 | gpt-5.6-sol (reasoning.mode: pro) |
o3-mini · o3-mini-2025-01-31 | gpt-5.6-sol |
o4-mini · o4-mini-2025-04-16 | gpt-5.6-terra |
gpt-image-1 | gpt-image-2 |
Fine-tuned variants: ft-gpt-3.5-turbo · ft-gpt-4 · ft-gpt-4.1-nano-2025-04-14 · ft-o4-mini-2025-04-16 · ft-babbage-002 · ft-davinci-002 | Base replacements in the gpt-5.6 family (re-run fine-tuning where needed) |
A third wave retires legacy audio/realtime models (gpt-4o-audio, gpt-realtime and others) on January 20, 2027.
How to audit your client workflows this week
- 01
n8n
Export each workflow as JSON and search the file for
gpt-3.5,gpt-4,o1,o3-mini,o4-mini. Check both OpenAI nodes and HTTP Request nodes callingapi.openai.com. - 02
Make
Open each scenario blueprint and check the model field in OpenAI modules — scenarios cloned from templates are the usual place old ids hide.
- 03
Zapier
Review ChatGPT/OpenAI actions in each Zap; anything configured before mid-2025 deserves a look.
- 04
Swap, run, diff
Switch to the replacement model, run the workflow once end-to-end, and diff the output — replacement models can change tone, length and JSON adherence, so test before the deadline, not after.
Don't want to do this manually for every client, every vendor?
This page covers one vendor's announcement. Your clients also depend on Shopify, Meta, Slack, Google and a dozen others — each with its own changelog. Get a free one-time dependency report for one workflow export, or let us watch all of it per client, continuously.