OpenAI pulls GPT-4o and GPT-4.1 from its API on January 14

OpenAI told developers on October 14 that it will remove five GPT-4 series model snapshots from its API on January 14, 2027. The list covers GPT-4o (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) and GPT-4.1 mini (gpt-4.1-mini-2025-04-14). For every one of them, OpenAI names gpt-6-luna as the replacement.

Many companies built chatbots, document extraction and classification jobs on these models in 2024 and 2025, and a lot of that code calls a fixed snapshot name. After January 14 those calls stop working. Fine-tuned models built on any of the five follow the same date, so a custom model trained on GPT-4o mini goes dark with its base.

A short notice period

OpenAI's deprecations page sets a minimum of six months' notice for generally available models, three months for specialized variants and as little as two weeks for previews, unless safety or compliance needs a faster exit. This notice gives three months. The October 14 entry does not give a reason.

The timing lands on top of other cutoffs already on the page. Older models announced for retirement in April, including the original gpt-4o-2024-05-13 snapshot, GPT-4.1 nano, o1, o3-mini and o4-mini, shut down on October 23, eight days from now. GPT-5 and o3 snapshots follow on December 11. The tts-1 and tts-1-hd speech models go on January 6.

OpenAI says some customers can keep access after a shutdown date by buying dedicated capacity through its sales team. It has not published what that costs.

The cost side

On list prices, the move should cut token bills. OpenAI's model pages, as compiled by CellCog, put the models side by side:

That makes Luna's input price one twenty-fifth of GPT-4o's. The table does not settle the question for every workload. Luna is a newer model with reasoning settings, and its answers can use more tokens than the model it replaces. OpenAI's own migration guidance tells teams to measure total cost on their own traffic.

The larger risk is behavior. A prompt tuned for GPT-4o can return a different format, call tools in a different order or refuse requests the old model answered. Anything that parses model output into a database, a ticket or a customer email needs testing before the switch.

This notice covers OpenAI's own API. Microsoft publishes separate retirement dates for the same models in Azure, so companies that buy through Azure should check that schedule as well. Our GPT-6 rollout story covers the ChatGPT side, and our look at Gemini 4 Argon pricing is the comparison point if you are weighing a second vendor.