GPT-4o and GPT-4.1 snapshots leave OpenAI's API on January 14. Find every process that still calls one

The automations most likely to break in January are the ones nobody has looked at since they went live. Picture an intake form a contractor wired up in 2025: it has worked ever since, and nobody has changed the model name inside it. If someone outside your company built that process, add one question to the ones in how to choose AI help and what to ask any vendor: who changes the model when the vendor retires it?

What OpenAI announced

On October 14, OpenAI notified developers that five named snapshots will be removed from its API on January 14, 2027. According to OpenAI’s deprecations page, they are:

gpt-4o-2024-08-06 and gpt-4o-2024-11-20 (GPT-4o)
gpt-4o-mini-2024-07-18 (GPT-4o mini)
gpt-4.1-2025-04-14 (GPT-4.1)
gpt-4.1-mini-2025-04-14 (GPT-4.1 mini)

The page lists gpt-6-luna as the recommended replacement for all five. Fine-tuned versions of these models follow the same shutdown schedule as their base models, so a custom model built on GPT-4o mini goes too.

Two details are worth reading closely.

First, the clock. The same page says OpenAI normally gives at least six months’ notice before retiring a generally available model, unless safety or compliance concerns require a faster timeline. This notice is three months. OpenAI’s entry does not say why.

Second, the scope. The entry names dated snapshots only. It does not mention the undated names such as gpt-4o or gpt-4.1-mini, and it does not say what a request to a removed model returns after the shutdown. Do not assume either way. Check what your code actually sends.

Two more dates on the same page

If any of your processes talk or listen, the same page lists two more deadlines close by:

January 6, 2027: the text-to-speech models tts-1, tts-1-hd and two gpt-4o-mini-tts snapshots are removed. The listed replacement is gpt-realtime-2.1-mini.
February 26, 2027: the transcription models whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize are removed. The listed replacements are gpt-live-transcribe or gpt-transcribe.

A voice-note-to-CRM process could depend on all three dates at once.

The test you can run this week: the model-name sweep

You are looking for every place your business names a model. Most of them are not in a code repository.

Search your code. In every repository you own, search for gpt-4o, gpt-4.1, tts-1, whisper-1 and transcribe. Note the file and what the process does.
Search your tools. Open Zapier, Make, n8n or whatever runs your automations. Open each AI step and write down the model it uses.
Search your settings. Look in environment variables, config files and spreadsheet formulas. Model names often hide in a setting called MODEL that nobody reads.
Ask your vendors. Email every contractor or agency that built something for you: “Which OpenAI models does what you built call, and who updates it before January 14?” Ask for the answer in writing.
Check fine-tunes. In your OpenAI account, list any fine-tuned models and note which base model each one was trained on.
Rank what you found. For each hit, write one line: what breaks if it stops, who notices, and how fast.

The result is a short table: process, model name, owner, impact. The number of rows is the size of the job.

A worked plan: 92 days, counted backward

January 14 is 92 days from October 14, and the holidays sit inside that window. A plan that leaves room for them:

By October 23: sweep done, every row has an owner.
By November 13: for each process, run 20 real past inputs through the replacement model and compare the outputs side by side with what the current model produced. Look at format as well as content, because downstream steps often parse the output.
By December 4: switch the processes that passed. Keep the old model name written down in case you need to compare later.
By December 18: fix the ones that did not pass, or decide to retire them.
December 18 to January 6: buffer. If you use text-to-speech, January 6 is your real deadline, not January 14.

Do not skip the 20-input comparison. A replacement model can be better on average and still change the shape of an answer that a later step depends on.

The decision rule

For each process in your table, ask one question: if this stopped on January 14, would anyone notice within a day?

Yes, and it matters: move it first and test it properly.
Yes, but it is minor: move it in the normal queue.
No: ask whether it should still exist. A process nobody would miss is a candidate to turn off, not to migrate.

What to do this week

Run the model-name sweep across code, automation tools, settings and fine-tunes.
Send the one-line vendor email to everyone who built something for you.
Put January 6, January 14 and February 26 on the calendar of whoever owns each process.
Pick 20 real past inputs for your most important process and save them for the comparison.

If you want a second pair of hands, a consult block picks the one or two workflows where AI saves the most time or money, works on them with you on your real material, and you leave with a written plan.