In this article
On October 14, 2026, OpenAI announced that five GPT-4 family models will be shut down in its API on January 14, 2027: both versions of GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini. Any chatbot, invoicing tool or message-sorting system a developer built for you on one of these models will stop answering that day unless it is moved first. OpenAI's recommended replacement is GPT-6 Luna, priced at $0.10 per million input tokens and $0.50 per million output tokens, against $2.50 and $10 for GPT-4o, roughly 20 to 25 times cheaper. You have 92 days: audit your systems this week, test the replacement on your own data, and switch before mid-December.
What exactly OpenAI announced
OpenAI published the notice on its developer deprecations page. It lists the following model IDs, all shutting down on January 14, 2027:
- gpt-4o-2024-08-06 and gpt-4o-2024-11-20 (GPT-4o)
- gpt-4o-mini-2024-07-18 (the small model many systems were built on because it was cheap)
- gpt-4.1-2025-04-14 and gpt-4.1-mini-2025-04-14
The page states that fine-tuned versions of these models follow the same shutdown date as their base model. The notice covers the API, meaning systems built by developers, not the ChatGPT app your staff use in the browser.
An earlier deadline: October 23, 2026
The same page carries an older notice, issued on April 22, 2026, that shuts down older models just 9 days from today, on October 23, 2026. They include the first GPT-4o snapshot gpt-4o-2024-05-13, gpt-4-turbo, gpt-3.5-turbo, gpt-4.1-nano, o1, o3-mini and o4-mini. If your system is old and has not been updated in two years, it may run on one of these, and that is worth a call to your developer today, not next month.
How your bill changes: a worked example
Prices below come from OpenAI's pricing page, in US dollars per million tokens (a token is roughly part of a word):
| Model | Input | Output |
|---|---|---|
| GPT-4o | $2.50 | $10.00 |
| GPT-4.1 | $2.00 | $8.00 |
| GPT-4.1 mini | $0.40 | $1.60 |
| GPT-4o mini | $0.15 | $0.60 |
| GPT-6 Luna (replacement) | $0.10 | $0.50 |
A hypothetical example for illustration: a customer service chatbot handling 20,000 conversations a month, each with about 3,000 input tokens and 500 output tokens. On GPT-4o it costs about $250 a month (around SAR 937 at 3.75), and on GPT-6 Luna about $11 (around SAR 41). If it runs on GPT-4o mini the gap is smaller: about $15 against $11. Note that Luna is a reasoning model with a default effort of medium, and that reasoning is billed as output tokens; set it to none or low for simple tasks, or the bill will come in above this example.
What can break when you switch
- Fine-tuned models: according to the GPT-6 Luna model page, it does not support fine-tuning. If your system depends on a fine-tuned GPT-4o, there is no ready equivalent, and you will need sharper instructions and examples instead, or another model.
- Tool calls: the same page says function calling through Chat Completions works with Luna only when reasoning effort is set to none. A system that connects AI to your inventory or booking system may stop carrying out actions if this is not set.
- Assistants API: Luna does not run on the Assistants or Realtime endpoints, so systems built on them need partial rebuilding, not just a new model name.
- Factual accuracy: in our earlier comparison Claude Haiku 5.5 vs GPT-6 Luna we cited Artificial Analysis testing showing that Luna often gets things wrong when answering from memory. If your chatbot answers questions about your prices and policies, give it the information in every request rather than relying on what it remembers.
Migration steps before January 14
- Audit your systems this week: ask your developer to search the code and settings for gpt-4o, gpt-4.1 and gpt-3.5, and check the Usage page in your OpenAI account to see which models are actually being called.
- Start with what stops on October 23: anything on the older models is top priority.
- Test on a real sample: take 100 to 200 real conversations or documents, run them through the new model, and compare the answers with the old one before switching.
- Set the reasoning effort: none for fast tasks and tool calls, higher for tasks that need analysis.
- Switch before mid-December: so you keep at least four weeks to catch any fault before the deadline.
If these systems process your customers' personal data, note that the Luna page lists data residency in EU regions and does not mention Saudi Arabia, so review the Personal Data Protection Law requirements before you move. To compare OpenAI's current models, read the difference between GPT-6 Astra, Sol and Luna.
The Origami view
Model shutdowns are now routine: OpenAI's page alone carries at least eight deprecation notices since April 2026. That is why we build our clients' systems so the model name is a single setting that changes without reprogramming, backed by a fixed test set that exposes any drop in quality within hours. If you have a system built by a developer who has since left and you do not know which model it runs on, get in touch and we will check it for you, or see our services.
Conclusion
January 14, 2027 is a hard deadline that is rarely extended, and October 23 is closer than it looks. The good news is that the replacement is far cheaper, but switching is not a one-word change in the code: test it on your data, set the reasoning and tool options, and move early.
Sources
Frequently asked questions
When does GPT-4o stop working in the OpenAI API?
On January 14, 2027, together with GPT-4o mini, GPT-4.1, GPT-4.1 mini and their fine-tuned versions, according to OpenAI's notice of October 14, 2026. The first GPT-4o snapshot, gpt-4o-2024-05-13, stops earlier, on October 23, 2026.
What replacement does OpenAI recommend, and what does it cost?
OpenAI recommends GPT-6 Luna for all five models. It costs $0.10 per million input tokens and $0.50 per million output tokens, against $2.50 and $10 for GPT-4o, roughly 20 to 25 times cheaper.
Is it enough for our developer to change the model name in the code?
Usually not. GPT-6 Luna does not support fine-tuning, does not run on the Assistants API, and tool calls through Chat Completions need reasoning effort set to none. Test it on 100 to 200 real cases before you switch.
Are staff who use ChatGPT in the browser affected?
No. The notice covers the API, meaning systems developers built for you such as chatbots or automation tools, not the ChatGPT app itself.
Follow Origami in Google
Pin Origami as a preferred source and our articles will surface first for you in Google Search and Top Stories.
