GPT-4o leaves the OpenAI API on 14 January. Your checklist
Key takeaways
- OpenAI will remove two GPT-4o snapshots, GPT-4o mini, GPT-4.1 and GPT-4.1 mini from its API on 14 January 2027.
- OpenAI recommends gpt-6-luna as the replacement for all of them, and fine-tuned versions follow the same shutdown date as their base models.
- Search your code and automations for the old model IDs now, test prompts on the replacement, and set a cut-over date well before the deadline.
13 weeks. That is how long developers have between OpenAI's deprecation notice on 14 October and the 14 January 2027 shutdown of five GPT-4 era models in its API. Plenty of small apps, Zapier flows and side projects still call these models without anyone remembering they do.
Which models are going
OpenAI's deprecations page lists five snapshots: gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14 and gpt-4.1-mini-2025-04-14. All five map to the same recommended replacement, gpt-6-luna. Fine-tuned versions follow the same schedule as their base models, so a custom model built on GPT-4o disappears on the same day.
This is separate from ChatGPT, where GPT-4o was retired earlier in 2026. The API date is the one that breaks code.
The migration checklist
- Search every repository, config file and environment variable for the five model IDs above. Do not forget automation tools and no-code flows, which often hide a model name in a dropdown.
- List any fine-tuned models and note which base model each one uses. They need retraining or replacing before 14 January.
- Run a sample of real prompts through gpt-6-luna and compare the answers side by side. Use actual inputs from your logs, not tidy test cases.
- Check output format. If your code parses JSON or expects a fixed structure, confirm the new model still returns it the same way, including on edge cases and refusals.
- Compare cost. OpenAI's deprecations page does not list prices for the replacement, so check the current pricing page before you commit to a budget.
- Pick a cut-over date in early December, leave the old model running as a fallback until the deadline, then remove it.
Whether to stay with OpenAI
A forced migration is a reasonable moment to price up alternatives. We have covered Gemini 4 Argon's pricing, Claude Haiku 5.5's pricing and the new Meta Muse Wonder model. Switching providers means rewriting more than a model name, so test before you decide.
The part worth watching is December. Retirement dates tend to break things quietly, and the apps that fail first are usually the ones nobody has touched in a year.
