6 min read
It is a Tuesday morning in mid January. A customer fills in the quote form on your website, and nothing happens. No instant reply, no draft in your inbox, no new row in your spreadsheet. The automation you set up two years ago, the one you have not thought about since, quietly stopped working overnight. Nobody emailed you about it, because the warning went out in October to an email address you never read.
That scenario is avoidable, and you have about three months to avoid it. On October 14, OpenAI announced that several older models will be removed from its API on January 14, 2027. This guide walks you through, step by step, how to find out whether anything in your business depends on them and how to switch safely. No coding needed.
What is actually being retired, in plain English
The API is the behind-the-scenes connection that lets other software (Zapier, a WordPress plugin, a website chatbot) send text to OpenAI’s models and get answers back. It is separate from the ChatGPT app you chat with. According to OpenAI’s API deprecations page, as summarized in developer coverage of the notice, the models leaving are:
- GPT-4o and GPT-4o mini
- GPT-4.1 and GPT-4.1 mini
- Any fine-tuned versions of those models (custom versions trained on your own examples), which follow the same shutdown date
OpenAI recommends GPT-6 Luna as the replacement for all four. Reports also say the notice applies only to the API. If you only ever type into ChatGPT itself, this change does not affect you directly. Older voice and real time audio models have their own, separate dates, so check those too if you run a phone or voice tool.
Step 1: Find out if you use the API at all
Many solo owners are not sure whether they have an API account. Here is the quick test.
- Search your email for “OpenAI API” or “platform.openai.com” and for receipts that mention “API usage” or “credits”.
- Log in at platform.openai.com with the same account. If you see a billing balance or API keys (long codes starting with “sk-“), you have an API account.
- If a freelancer or agency built your chatbot or automations, ask them directly: “Does anything you built for me call OpenAI, and which model does it use?”
If none of these turn anything up, you are probably fine. Keep reading anyway, because some tools use OpenAI through their own account and will handle the switch for you (more on that in step 3).
Step 2: Read your usage page to see which models you actually call
This is the single most useful five minutes in this guide. In the OpenAI platform dashboard, open the Usage section and look at the breakdown by model for the last 30 days. Every model name that shows up there is something running in your business.
Write down any line that includes gpt-4o, gpt-4o-mini, gpt-4.1 or gpt-4.1-mini, including longer names with dates on the end, such as gpt-4o-2024-08-06. Those are your to-do list. If you see none of them, you are already on newer models and can stop here.
Step 3: Check the three places small businesses usually hide these models
Now match each model on your list to the tool that uses it. In our experience, solo businesses almost always find them in one of three places.
Automation tools like Zapier and Make
Open each automation (a Zap in Zapier, a scenario in Make) that has an OpenAI or ChatGPT step. Inside that step there is a Model dropdown. If it says gpt-4o or gpt-4.1, write down the automation’s name. If you are rethinking your automation setup anyway, our Zapier vs Make vs n8n comparison can help you decide whether to fix it in place or move.
Website plugins and chatbots
AI writing plugins, chat widgets and form tools that ask you to paste in an OpenAI key usually have a settings screen with a model choice. Check each one. If a plugin does not let you choose, check whether it has had an update recently; well maintained plugins will switch models for you before the deadline.
Tools that use their own OpenAI account
Many apps (email tools, CRMs, design apps) include AI features without asking for your key. They pay OpenAI themselves, so it is their job to migrate. You do not need to do anything, but it is worth skimming their changelog in December to be sure.
Step 4: Test the replacement on your real work before switching
Do not flip every switch at once. Newer models are generally better, but “better” can also mean longer answers, a different tone, or a changed format that breaks the next step of an automation. Here is a gentle way to test.
- Pick five real examples from each automation: five past customer inquiries, five product descriptions, whatever that automation handles.
- Duplicate the automation (Zapier and Make both let you copy one), and set the copy’s model to the recommended replacement.
- Run your five examples through both versions and put the outputs side by side.
- Check three things: is the tone right, is the length right, and does the next step still work (for example, does the spreadsheet still get the right columns).
If the new output is too long, add a line to your instructions such as “Reply in no more than 80 words.” That one sentence fixes most surprises.
Step 5: Compare the cost, then switch
Price per use differs between models, so look at the replacement’s price on OpenAI’s pricing page and multiply by your monthly usage from step 2. For most solo businesses, the numbers are small either way. While you are comparing, it is also a reasonable moment to ask whether a cheaper model from another provider would do the job; our roundup of Anthropic’s October updates for solo owners covers one low cost alternative.
Once the test copy looks right, switch the original automation over, turn off the copy, and run one live example to confirm.
Step 6: Leave yourself a note so this never surprises you again
Models will keep being retired. The fix is a tiny habit, not more software.
- Keep a simple list (a note or spreadsheet) of every automation that uses AI, which model it uses, and who built it.
- Make sure the email address on your OpenAI account is one you actually read.
- Put a calendar reminder for December 15, 2026 to recheck your usage page. If a retiring model still shows up, you have a month to fix it.
If you have fine-tuned models, start sooner: one report says new fine-tuning jobs on the retiring models stop being accepted on January 6, 2027, a week before the shutdown.
A calm checklist for the next three months
- This week: steps 1 and 2. Find out if you are affected.
- By mid November: step 3. List every tool that uses a retiring model.
- By mid December: steps 4 and 5. Test, then switch.
- December 15: step 6. Recheck and relax.
You do not need to be technical to do any of this. You need one quiet hour and the usage page. If you have been wondering whether to rely less on outside AI services altogether, our guide to running a private AI model on your own laptop is a good next read.
Did your usage page turn up anything you had forgotten about? Share what you found in the comments; it might save another reader from that January surprise.
Related reading
- Zapier vs Make vs n8n: Which Automation Tool Should a One-Person Business Actually Pay For?
- Claude Haiku 5.5, Cheaper Sonnet and Monthly API Credits: Anthropic’s October Updates for Solo Owners
- How to Run a Private AI Model on Your Own Laptop for Client Work You Should Not Upload
- GPT-6 in ChatGPT Review: What Free Luna and Paid Sol Mean for a Solo Business
