Published: 15 October 2026
OpenAI told developers on 14 October that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from its API on 14 January 2027. From that day, apps and automations can no longer call these models, and fine-tuned versions stop at the same time. OpenAI recommends GPT-6 Luna as the replacement for all four.
The notice names five model versions. If your code or settings contain one of these names, it is affected:
- gpt-4o-2024-08-06 and gpt-4o-2024-11-20
- gpt-4o-mini-2024-07-18
- gpt-4.1-2025-04-14
- gpt-4.1-mini-2025-04-14
Two older versions have even less time. Under an earlier notice from April, gpt-4o-2024-05-13 and gpt-4.1-nano leave the API on 23 October 2026, eight days from now.
ChatGPT users lost these models in February, now API integrations follow
If you only use ChatGPT, nothing changes for you. OpenAI removed GPT-4o, GPT-4.1 and GPT-4.1 mini from ChatGPT on 13 February 2026 and said at the time that the API would stay as it was. This notice is about the API: internal tools, customer chatbots, document pipelines and no-code workflows that send requests to OpenAI with one of these model names.
For most organisations, the hard part is not the code. Changing a model name takes a minute. Finding every place where a name is set takes longer. It can sit in a Make or Power Automate flow, a plugin on your website or a script a former colleague wrote. It can also sit in a supplier’s product that runs on OpenAI behind the scenes. Then each of those places needs a test, and the three months include the December holidays.
Every model on the list costs more per token than GPT-6 Luna
Even GPT-4o mini, the cheapest of the four, has a higher list price than its replacement.
| Model | Input per 1M tokens | Output per 1M tokens | Context window |
|---|---|---|---|
| GPT-4o | $2.50 | $10.00 | 128,000 tokens |
| GPT-4o mini | $0.15 | $0.60 | 128,000 tokens |
| GPT-4.1 | $2.00 | $8.00 | about 1 million tokens |
| GPT-4.1 mini | $0.40 | $1.60 | about 1 million tokens |
| GPT-6 Luna | $0.10 | $0.50 | about 1 million tokens |
These are OpenAI’s standard list prices from its model pages. They do not tell you the cost per task, because GPT-6 Luna’s reasoning settings change how many tokens a request uses. OpenAI itself advises comparing total cost on your own workload.
GPT-6 Luna reasons by default, which GPT-4o and GPT-4.1 never did
GPT-4o and GPT-4.1 answer straight away. GPT-6 Luna can reason first, and its default reasoning effort is “medium”. OpenAI’s migration guide advises starting with the effort set to “none”, which it expects to suit most GPT-4.x use cases. Raise it only where your tests show that quality improves enough to pay for the extra tokens and waiting time.
One detail for developers: in the older Chat Completions API, GPT-6 Luna supports function calling only with reasoning set to “none”. For built-in tools such as web search or file search, OpenAI points to the Responses API.
Fine-tuned GPT-4o and GPT-4.1 models have no direct successor
This is the part of the notice with the biggest consequences. Fine-tuned versions of the four models stop on 14 January as well, and GPT-6 Luna cannot be fine-tuned. OpenAI is also winding down its self-serve fine-tuning platform, and active customers can start new training jobs only until 6 January 2027.
If you trained a model on your own examples, plan to rebuild that behaviour another way. Start with clear instructions and a handful of your training examples in the prompt. Use retrieval from your own documents for knowledge the model has to look up. Where neither is good enough, the task may need a provider that still offers fine-tuning.
Does the 14 January date apply to Azure OpenAI?
No, Microsoft runs its own schedule. Its model retirement page, last updated on 23 September, lists 14 April 2027 for the same GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini versions. Microsoft names GPT-5.1 as the replacement for GPT-4o and does not list GPT-6 Luna yet. Check that page again before you plan around the extra three months, because it can change.
If you call OpenAI directly and need your data processed in Europe, GPT-6 Luna supports OpenAI’s EU data residency. OpenAI adds 10% to the price for regional processing on models released on or after 5 March 2026, and GPT-6 Luna is one of them.
Our advice is to start with an inventory this month. List every place that calls one of these model names, including no-code tools, and ask suppliers which model they will move to and when. Then follow the order OpenAI suggests. Record today’s quality, speed and cost on real requests, test GPT-6 Luna with reasoning off against that baseline, and move traffic over in steps. A forced switch is also a good moment to compare GPT-6 Luna with similar-priced models from other providers, such as Claude Haiku 5.5. While you are at it, make the model name a setting instead of something written into your code. Aim to finish before Christmas, not in the second week of January.
For more information, visit the official announcement of the GPT-4o and GPT-4.1 shutdown on the OpenAI API deprecations page.
