Blog
OpenAI will remove GPT-4o and GPT-4.1 from its API on 14 January 2027
On 14 October 2026 OpenAI told developers that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from its API on 14 January 2027, along with fine-tuned versions. It recommends GPT-6 Luna as the replacement. ChatGPT users are not affected.
OpenAI has set a shutdown date for some of its best-known older models. On 14 October 2026 it notified developers that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from its API on 14 January 2027. After that date, apps that call these models by name will stop working.
The notice is listed on OpenAI's API deprecations page. It applies to the API only. People who use ChatGPT do not need to do anything.
Which models are affected
The deprecation names five specific model versions (snapshots):
- GPT-4o:
gpt-4o-2024-08-06andgpt-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
All five shut down on 14 January 2027. Fine-tuned versions of these models follow the same schedule, so custom models built on them also stop on 14 January.
For all four, OpenAI recommends `gpt-6-luna` as the replacement.
Some older versions go sooner
A separate notice from April already covers other GPT-4-era models, which shut down on 23 October 2026. These include gpt-4o-2024-05-13, gpt-4.1-nano, gpt-4-turbo and gpt-4-0613. Developers still pinned to those have days, not months.
What switching to GPT-6 Luna involves
According to OpenAI's model page, GPT-6 Luna is its "most efficient model for focused, high-volume tasks". A few details matter for anyone migrating:
- Price: $0.10 per million input tokens and $0.50 per million output tokens at standard rates.
- It reasons by default. Its default reasoning effort is
medium, which uses extra tokens and time. It can be set as low asnone. - Function calling: in the older Chat Completions API, GPT-6 Luna supports function calling only when reasoning effort is set to
none. OpenAI points developers to the Responses API for built-in tools. - No fine-tuning: the model page lists fine-tuning as not supported.
That last point is the hardest for some teams. OpenAI is also winding down self-serve fine-tuning: according to the same deprecations page, existing customers can create new fine-tuning jobs only until 6 January 2027. Teams that rely on a fine-tuned GPT-4o or GPT-4.1 have no direct fine-tuned successor on OpenAI's platform and will need to rebuild that behaviour another way, for example with instructions and examples in the prompt.
About the notice period
OpenAI's deprecations page says it normally gives generally available models at least six months' notice, unless safety or compliance concerns require a faster timeline. This notice gives three months. OpenAI's page does not explain the shorter period.
What developers should do
- Search code and settings for the snapshot IDs above.
- Check the
shutdown_datefield returned by OpenAI's List models endpoint, which OpenAI says shows shutdown dates programmatically. - Test GPT-6 Luna on real workloads, starting with low reasoning effort, before switching.
- Plan early for any fine-tuned models.
Azure OpenAI runs its own retirement schedule, so customers using these models through Microsoft should check Microsoft's documentation separately.