On October 14, 2026, OpenAI told developers that four GPT-4 models will be removed from its API on January 14, 2027. The notice is on OpenAI's deprecations page. The four are GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini. Fine-tuned versions of these models stop on the same date. OpenAI names gpt-6-luna as the replacement.
What exactly changes
The API is the connection software uses to call a language model from OpenAI. A program asks for a model by a fixed name, for example gpt-4o or gpt-4o-2024-08-06. After January 14, 2027, the four models on the list no longer exist in that API. Software that still asks for one of them will no longer get an answer from the model.
The notice is about the API, so about software that calls a model through that connection. That is software someone built or set up for you, or that you set up yourself.
What it means for a business owner
This is not about which model is newest or smartest. It is about things that can quietly stop working.
Think of a chatbot on your website, a script that sorts your inbox, a connection that writes quotes or summaries, or an automation someone built for you two years ago. Any of them can call one of these four models by name. You cannot see that from the outside. The chatbot looks the same and the script runs in the background, until one day the call fails.
The hard part is the timing. A failure like that shows up when a customer gets no reply, or when a report suddenly stays empty. Then you first have to find out what went wrong, and only after that can you fix it.
What you can do this week
Make a short list. Three questions per tool:
- 1Which tools in your business use AI? Include things that have been running for a long time and that you no longer think about, such as a form that summarises answers or an email that is drafted automatically.
- 2Who set it up? An employee, a freelancer, an agency, or you yourself with an off-the-shelf package.
- 3Which model does it call? If you do not know, ask the person who built it. They can look it up in the settings or in the code.
If one of the four models is on your list, plan a moment to switch now, well before January 14. Changing the model is usually small work. Do test afterwards whether the answers still do what you expect: a different model can phrase things a little differently. And if you had a fine-tuned version made, filling in a new name is not enough. That version stops on the same date and has to be made again on a model that stays.
Note: an older version of GPT-4o (gpt-4o-2024-05-13) and GPT-4.1 nano already stop on October 23, 2026. They are on the same page, with their own replacement.
If you use tools from a software vendor, ask them whether they are already handling this. Then you know for sure and you do not have to wait until January.
Why this will come back
The same page lists other models with a shutdown date too, with the most recent notice at the top. So an AI tool is not something you build once and then forget. If you know what runs in your business and which model it uses, the next notice is a short job instead of a surprise.
When you have a new connection built, ask for a short description with it: what it does, which model it uses and who maintains it. You can read more about how such a connection works in Having an API integration built.
Not sure what runs on which model in your business? Send us a message and we will check it with you.
