OpenAI retires GPT-4o and GPT-4.1 from its API on 14 January 2027. Will your automations break?

Is OpenAI shutting down GPT-4o in the API?

Yes. On 14 October 2026 OpenAI added an entry to its API deprecations page: GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini shut down on 14 January 2027. OpenAI recommends one replacement for all four, gpt-6-luna, and notified developers using these models on the same day.

OpenAI adds that fine-tuned versions of these models follow the same schedule. The same page also retires the tts-1, tts-1-hd and gpt-4o-mini-tts voice models on 6 January 2027, and gpt-5.1, gpt-5.3-codex and gpt-5.4-nano on 1 April 2027.

This is an API change. It affects anything that calls OpenAI with your API key: Zapier, Make and n8n steps, website chat widgets, WordPress and CRM plugins, and scripts a developer wrote for you.

Will my Zapier, Make or n8n automations stop working?

Any step still set to one of these models will fail from 14 January. There is a precedent. When OpenAI removed gpt-4-vision-preview in 2024, Zapier users posted a "model_not_found" error saying the model "has been deprecated", and the fix was to move the step to a newer model.

The hard part is knowing where these models are. For two years GPT-4o was the safe default, so many workflows were built on it and never touched again. Make's OpenAI help page still names GPT-4o and GPT-4.1 as non-reasoning models you can switch its AI agents to. n8n loads the model list from your OpenAI account, so a workflow keeps whatever model was picked when it was built.

How do I find every place my business uses GPT-4o or GPT-4.1?

Use this checklist this week:

1. 1Search the inbox of whoever owns the OpenAI account for OpenAI's 14 October notice.
2. 2Check usage in your OpenAI account and note which models are still being called.
3. 3Open every automation with an OpenAI step in Zapier, Make and n8n, and write down the model each one uses.
4. 4Search your code and settings for "gpt-4o" and "gpt-4.1", including environment variables and config files.
5. 5List third-party tools that hold your API key, such as chat widgets, plugins and CRM add-ons, and check their model setting.
6. 6For developers: OpenAI suggests reading the shutdown_date field from its List models endpoint to track shutdowns in code.

If the list runs past a handful, our AI workflows service starts with exactly this kind of workflow map, and can take the switch off your plate.

Is GPT-6 Luna a drop-in replacement?

Often, but not always. Here is how OpenAI's own model pages compare them (read on 3 November 2026, prices per 1 million text tokens):

Three things to test before you switch:

• Reasoning settings. GPT-6 Luna has a reasoning effort setting that defaults to medium. OpenAI's page says that in the Chat Completions API, function calling only works when reasoning effort is set to none. A step that calls tools may need that setting changed, or a move to the Responses API.
• Fine-tuned models. These stop on 14 January too, and Luna cannot be fine-tuned. Our advice: a prompt with a few good examples can cover simple sorting and tagging jobs, but test it.
• Tone. A new model writes differently. Anything that goes out in your brand voice, such as email replies or product descriptions, needs a side-by-side read.

Our advice: start with Luna for sorting, summarising and data clean-up, because the list price is lower than all four models it replaces. Watch the first week's bill and outputs, not just the price page.

Worked example: a six-person agency with four AI automations

This is a hypothetical scenario, not a client.

1. 1Day 1: inventory. The operations lead finds four uses: a Zap that sorts website enquiries with GPT-4o mini, an n8n workflow that drafts email replies with GPT-4o, a Make scenario that summarises sales calls with GPT-4.1, and a chat widget on the website set to GPT-4o mini in its plugin settings.
2. 2Week 1: copy and test. She duplicates each automation, switches the copy to gpt-6-luna and runs 20 real past inputs through old and new versions.
3. 3Week 1: fix the gotchas. The reply drafter uses a tool call, so its reasoning effort is set to none. The call summaries come back longer, so the prompt gets a word limit.
4. 4Week 2: brand check. The account lead reads ten drafted replies side by side and adds three approved example replies to the prompt.
5. 5Before mid-December: switch. Live steps move to the new model. The old versions stay paused, not deleted, for a week.
6. 6Two weeks after: monitor. She checks error logs and the OpenAI bill, then deletes the old copies.

The order matters more than the speed. The inventory is the step most teams skip, and it is the one that prevents a silent failure on 14 January.

Should you switch now or wait until January?

Our advice is to switch now. 14 January falls straight after the holidays, when the person who built the workflow may be away, and a failed lead-sorting step can go unnoticed for days. It is not the only automation deadline this season: see our guides to OpenAI Agent Builder alternatives before 30 November and Pipedream Workflows alternatives. While you are in there, ask whether each automation still earns its place.

Can Theepa help?

Yes. Our AI workflows and automation service starts with a map of your workflows and ends with a setup your team can run. If you are not sure the website those automations feed is working either, start with a free brand audit.

Sources

• Deprecations, entry of 14 October 2026 (OpenAI API docs, accessed 3 November 2026)
• GPT-6 Luna model page (OpenAI API docs, accessed 3 November 2026)
• GPT-4o model page (OpenAI API docs, accessed 3 November 2026)
• GPT-4o mini model page (OpenAI API docs, accessed 3 November 2026)
• GPT-4.1 model page (OpenAI API docs, accessed 3 November 2026)
• OpenAI app help (Make, accessed 3 November 2026)
• OpenAI Chat Model node (n8n Docs, accessed 3 November 2026)
• Model not found: gpt-4-vision-preview has been deprecated (Zapier Community, 19 June 2024)