OpenAI retires GPT-4o and GPT-4.1 on January 14, 2027

OpenAI posted a deprecation notice today: the GPT-4o and GPT-4.1 model snapshots shut off from the API on January 14, 2027. That is three months, and it lands right after the holiday rush. If your business runs an AI workflow that was built in 2024 or 2025, there is a good chance it calls gpt-4o-mini or gpt-4.1-mini. Those were the cheap, fast defaults. They were in every tutorial, every Zapier template, and every "quick script" somebody wrote on a Friday.

What actually happened

Per OpenAI's deprecations page, the notice is dated 2026-10-14 and the removal date is 2027-01-14. Five snapshots are on the list:

• 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

The recommended replacement for all of them is gpt-6-luna. One line in the notice matters more than the rest: fine-tuned versions of these models follow the same shutdown schedule as their base models. If you paid to fine-tune gpt-4o-mini on your support tickets or product copy, that fine-tune stops working on January 14 too.

This is the second batch in two weeks. On October 1, OpenAI also scheduled gpt-5.1, gpt-5.3-codex, and gpt-5.4-nano for removal on April 1, 2027, and four text-to-speech models for January 6, 2027.

Why the GPT-4o deprecation matters for your business

The model-name swap takes five minutes. Finding every place the name lives takes longer.

Inventory first. Grep your repos for gpt-4o and gpt-4.1. Then check the places that are not in git: n8n and Make scenarios, Zapier AI steps, Google Apps Scripts, cron jobs, and vendor tools where you pasted an API key and picked a model from a dropdown. Search for the base name, not only the dated string. If you call an undated alias like gpt-4o, confirm in OpenAI's model docs which snapshot it points to and what happens to it on the 14th.

Re-test on your own data. A new model writes differently. Prompts tuned for 4o-mini can produce longer output, a different JSON shape, or new refusals on edge cases. Pull 30 real inputs — order emails, product descriptions, support tickets — and run them through both models. Diff the outputs. Check every parser that reads the response.

Plan the fine-tune separately. A fine-tuned model is not a string change. You need the training data, a target model that supports your use case, and a new eval. Start that now, not in January.

Price it. Run your real monthly token volume against Luna's pricing before you commit. A forced migration is a good time to ask whether OpenAI is still the right vendor for each job.

Sources: OpenAI API deprecations.