What happened
On October 14, 2026, OpenAI added four widely used models to its API deprecation schedule. GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from the API on January 14, 2027, giving developers three months to migrate. OpenAI recommends gpt-6-luna as the replacement for all four.
What’s new
| Model (snapshots) | Shutdown | Recommended replacement |
|---|---|---|
| gpt-4o (gpt-4o-2024-08-06, gpt-4o-2024-11-20) | Jan 14, 2027 | gpt-6-luna |
| gpt-4o-mini (gpt-4o-mini-2024-07-18) | Jan 14, 2027 | gpt-6-luna |
| gpt-4.1 (gpt-4.1-2025-04-14) | Jan 14, 2027 | gpt-6-luna |
| gpt-4.1-mini (gpt-4.1-mini-2025-04-14) | Jan 14, 2027 | gpt-6-luna |
Fine-tuned models built on these bases follow the same shutdown schedule as their base models.
This follows an earlier batch announced on October 1, 2026:
- gpt-5.3-codex and gpt-5.1 shut down April 1, 2027 (replacement: gpt-6-sol).
- gpt-5.4-nano shuts down April 1, 2027 (replacement: gpt-6-luna).
- tts-1, tts-1-hd and two gpt-4o-mini-tts snapshots shut down January 6, 2027 (replacement: gpt-realtime-2.1-mini).
Why it matters
GPT-4o and GPT-4o mini have been default choices in many production apps, tutorials, no-code tools and plugins since 2024. Any code that pins these model IDs will start failing after the shutdown date. The fine-tuning clause is the harder part: a custom model trained on GPT-4o or GPT-4.1 cannot simply be pointed at a new name, so teams need time to re-run fine-tuning or replace it with prompting on a newer model and re-evaluate quality.
The recommended replacement is also a newer model generation, not a renamed snapshot. Output style, length, tool-calling behavior and tokenization can differ, so prompts tuned for GPT-4o may need adjustment. Structured-output schemas and function definitions should be re-tested rather than assumed to behave identically. Cost can move in either direction depending on how many tokens the new model uses per task.
Third-party products that call the OpenAI API on your behalf, such as chat widgets, CMS plugins or support bots, may also be using these models. It is worth asking vendors which model they call.
How to prepare
- Search your codebase and configuration for gpt-4o and gpt-4.1 model strings, including environment variables and plugin settings.
- Run your evaluation set against gpt-6-luna (and gpt-6-sol for harder tasks) and compare quality, latency and cost.
- List any fine-tuned models and plan re-training on a supported base well before January 14, 2027.
- Check the OpenAI deprecations page periodically; it is the authoritative list.