Say your company runs a customer chatbot on its website, or a tool that reads supplier invoices and keys the data into your accounting system. Maybe your own team built it, or a contractor handed it over last year. Every day it quietly calls an OpenAI model through the API.
Then one day the model it calls is switched off. The chatbot stops answering, invoices pile up in a queue, and nobody in the company knows where the fix needs to go.
For several models that businesses use heavily, that day now has a date.
What OpenAI announced
On 14 October 2026, OpenAI notified developers using certain GPT-4 series models that they will be removed from the API on 14 January 2027. The models in this notice are:
- GPT-4o (snapshots gpt-4o-2024-08-06 and gpt-4o-2024-11-20)
- GPT-4o mini (gpt-4o-mini-2024-07-18)
- GPT-4.1 (gpt-4.1-2025-04-14)
- GPT-4.1 mini (gpt-4.1-mini-2025-04-14)
OpenAI recommends gpt-6-luna as the replacement for all four, and says fine-tuned versions of these models follow the same shutdown schedule as their base models.
That is about three months from notice to shutdown, and the window includes the year-end holidays when many teams are away.
Another set has less than ten days left
The same page carries an earlier notice from 22 April 2026 that comes due on 23 October 2026. Models shutting down that day include:
- gpt-3.5-turbo, gpt-4 and gpt-4-turbo
- gpt-4.1-nano and gpt-4o-2024-05-13
- o1, o1-pro, o3-mini and o4-mini
- gpt-image-1 for image generation
- Some fine-tuned models, such as ft-gpt-3.5-turbo and ft-gpt-4.1-nano-2025-04-14
Each one has its own suggested substitute, such as gpt-5.6-sol, gpt-5.6-luna or gpt-5.6-terra. If your systems still call any of these names, the work has to happen next week.
Why this is a business problem, not only an IT one
When a system is delivered, the model name usually sits in one line of code or a config file. Executives never see it and it is not in any IT asset register. Yet that one line can take the whole system down.
Switching models is also more than editing a name. A newer model may write longer or shorter answers, format output differently, read figures in documents differently, or cost a different amount per call. Any system that feeds AI output straight into an ERP or accounting system needs testing on real documents before the switch.
OpenAI says as much on its deprecations page: software that relies on its models may need occasional updates to keep working. The page also sets out minimum notice periods. Preview models, the ones with "preview" in their name, may be retired with much shorter notice, such as two weeks. That belongs in the decision when you pick a model for a business-critical system.
A checklist for businesses
- List every system that calls AI through an API, including ones built by contractors or provided by SaaS vendors, and note which model each one uses.
- Compare that list with OpenAI's Deprecations page, starting with the set that shuts down on 23 October 2026.
- Ask your contractor or vendor directly which model the system they delivered uses, when they plan to migrate, and who pays for the change.
- Build a test set from real work, for example 30 invoices or your most common customer questions, and compare the old and new model on it before switching.
- Have your systems check the shutdown_date field from the List models API on a schedule, so you know ahead of time without waiting for an email.
- Move the model name into a setting that can change without a code change, and name the person responsible for it.
If you cannot answer item 1, start there. Everything else depends on it.
What Enersys does
We have built and maintained business software for 14 years, including work that connects AI to business workflows and ERP systems. If you are not sure which of your systems call a model that is about to be retired, or you want help planning tests before a model switch, talk to the Enersys team.
Sources
