What happened
On 14 October OpenAI wrote to developers using its API with a date: on 14 January 2027, GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini stop working. The affected versions are gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14 and gpt-4.1-mini-2025-04-14.
The replacement it suggests for all of them is the same: GPT-6 Luna, the cheap model we already covered in our article on what automating with AI costs today.
One detail has annoyed a lot of people. OpenAI's own page says generally available models are retired with at least six months of notice. This time it was 92 days.
Why it matters more than it seems
For a year and a half, GPT-4o and GPT-4o mini were the default choice. If someone built you a chatbot, a workflow in n8n, Make or Zapier or an integration with your CRM between 2024 and early 2026, they most likely picked one of these two.
And those things don't warn you when they fail. On 14 January the API call returns an error, the workflow stops and the customer who messages you on WhatsApp at eleven at night gets no answer. You find out days later, when someone asks why no leads are coming in.
Watch out for what comes first, too: gpt-4o-2024-05-13 and gpt-4.1-nano are switched off this Friday, 23 October. That retirement was announced in April and plenty of people missed it.
What it means for a small business
The good news is that the switch usually works out cheaper. These are the official prices per million tokens:
- GPT-4o: 2.50 dollars input and 10 output.
- GPT-4.1: 2 dollars input and 8 output.
- GPT-4.1 mini: 0.40 input and 1.60 output.
- GPT-4o mini: 0.15 input and 0.60 output.
- GPT-6 Luna: 0.10 input and 0.50 output, with a 1,050,000 token context window.
The small print on GPT-6 Luna
Changing the model name in the settings and forgetting about it is not a good idea. There are three differences we have seen break things:
It reasons by default. GPT-6 Luna thinks before answering at "medium" effort unless you tell it otherwise. Classifying an email or pulling a figure from an invoice doesn't need that, and it adds response time and output tokens. It supports the "none" level, which is what we would use for almost every simple task.
It doesn't support fine-tuning. If you have a version of GPT-4o or GPT-4.1 tuned on your data, it also goes dark on 14 January and there is no direct equivalent in Luna. You'll need to rebuild that knowledge with good instructions, examples in the prompt or search over your documents.
It answers differently. Same prompt, different model, different tone and format. If your automation expects a JSON with specific fields or text with a certain structure, check it with real cases before switching.
And if you use Azure or a tool with AI built in
If your company uses these models through Azure OpenAI, Microsoft runs its own schedule: there, GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini are retired on 14 April 2027. You get three more months, but the work is the same.
If the AI comes inside software you pay for (a third-party chatbot, a plugin, a CRM), the migration is the vendor's job. Even so, we would ask them in writing which model they use and when they are switching, because a model change can alter how your assistant talks to your customers.
What I would do this week
There's no need to wait until December. One afternoon is enough to know what you have and what you're risking:
First, take inventory. Search your n8n, Make or Zapier workflows, your integration code and your configuration variables for any mention of gpt-4o, gpt-4o-mini, gpt-4.1 or gpt-4.1-mini. The usage dashboard on the OpenAI platform shows which models your account has used this month.
Second, check Friday's deadline. If gpt-4.1-nano or gpt-4o-2024-05-13 shows up, that's the urgent one.
Third, test with real cases. Take twenty or thirty examples of what each automation does, run them through GPT-6 Luna with reasoning set to "none" and compare. If something fails, try more reasoning effort or a different model.
Fourth, switch in December at the latest, not on 13 January. Leave room to see a couple of weeks of real use with the new model.
And a lesson for whatever you build from now on: the model name should live in one place, easy to change, and someone should keep a note of which model each process uses. It's exactly the kind of thing we review when we act as a company's fractional CTO.
If you want us to look at it with you
We design and maintain AI agents and automations for small and mid-sized businesses, and many of the ones we inherit from other providers run on GPT-4o mini. If your sales process runs through a CRM, we also build GoHighLevel automations.
Book a thirty-minute call. We go through which of your automations depend on the retiring models, which migration order makes sense and how much you save with the switch. You leave with a dated plan before January arrives. No commitment.
Sources
This article draws on the following publications, consulted on 19 October 2026:
Frequently asked questions
GPT-4o in its 2024-08-06 and 2024-11-20 versions, GPT-4o mini (2024-07-18), GPT-4.1 (2025-04-14) and GPT-4.1 mini (2025-04-14). Fine-tuned versions of those models are switched off on the same day. OpenAI recommends GPT-6 Luna as the replacement for all of them.
This notice is about the API, meaning the apps, chatbots and automations that call OpenAI from their own code or from tools such as n8n, Make or Zapier. People who use ChatGPT on the web or in the app don't need to do anything about this change.
Check the settings of your automation workflows and integrations for gpt-4o, gpt-4o-mini, gpt-4.1 or gpt-4.1-mini, and look at the usage dashboard of your account on the OpenAI platform, which shows consumption by model. If the AI comes inside third-party software, ask the vendor which model they use and when they will switch.
No. According to OpenAI's official pricing, GPT-6 Luna costs 0.10 dollars per million input tokens and 0.50 per million output tokens, compared with 2.50 and 10 dollars for GPT-4o and 0.15 and 0.60 for GPT-4o mini. Bear in mind that it reasons at medium effort by default, which produces more output tokens; for simple tasks it is worth setting it to none.
Microsoft has its own schedule. On Azure OpenAI, GPT-4o (2024-08-06 and 2024-11-20), GPT-4o mini, GPT-4.1 and GPT-4.1 mini are retired on 14 April 2027, three months later than on the OpenAI API.
Fine-tuned versions of GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini stop working on 14 January 2027, and GPT-6 Luna does not support fine-tuning. The usual approach is to replace that tuning with more precise instructions, examples in the prompt itself or search over your documents, and to test it with real cases before switching.