OpenAI removes its GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini API models on 14 January 2027. It named gpt-6-luna as the replacement for all of them, and fine-tuned versions stop on the same date.
You have about three months from the 14 October 2026 notice. That is shorter than the six months the same page promises for generally available models, so start the inventory now.
The short answer
- Notice: OpenAI notified developers on 14 October 2026.
- Removal date: 14 January 2027, about three months later. The page’s general policy says generally available models receive at least six months’ notice, and it does not explain the gap.
- Replacement: gpt-6-luna. The entry gives no detailed migration steps.
- Fine-tuned models: same shutdown schedule as their base models.
- Scope: the API only. The entry does not describe ChatGPT.
Which model IDs are being removed?
The listed IDs are 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.
| Model | Model IDs | Shutdown date | Recommended replacement |
|---|---|---|---|
| GPT-4o | gpt-4o-2024-08-06, gpt-4o-2024-11-20 | 14 January 2027 | gpt-6-luna |
| GPT-4o mini | gpt-4o-mini-2024-07-18 | 14 January 2027 | gpt-6-luna |
| GPT-4.1 | gpt-4.1-2025-04-14 | 14 January 2027 | gpt-6-luna |
| GPT-4.1 mini | gpt-4.1-mini-2025-04-14 | 14 January 2027 | gpt-6-luna |
Source: OpenAI API deprecations page, entry dated 14 October 2026.
How can you check your own models?
OpenAI advises reading each model’s shutdown_date field. Use the List models endpoint (GET /v1/models) or the Retrieve a model endpoint.
Fine-tuned versions of these models follow the same schedule as their base models, so they also stop on 14 January 2027. Retraining on a new base is the part that takes longest, so start there.
What else is being retired on the same page?
If you use more than one OpenAI API product, check these too. Several shutdowns land within months of each other.
| Announced | Models | Shutdown | Replacement |
|---|---|---|---|
| 1 October 2026 | gpt-5.3-codex, gpt-5.1 | 1 April 2027 | gpt-6-sol |
| 1 October 2026 | gpt-5.4-nano | 1 April 2027 | gpt-6-luna |
| 1 October 2026 | tts-1, tts-1-hd, gpt-4o-mini-tts-2025-03-20, gpt-4o-mini-tts-2025-12-15 | 6 January 2027 | gpt-realtime-2.1-mini |
| 26 August 2026 | whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, gpt-4o-transcribe-diarize | 26 February 2027 | gpt-live-transcribe or gpt-transcribe |
The 26 August 2026 entry also retires the /v1/audio/translations endpoint on 26 February 2027, with v1/realtime/translations as the replacement. The text-to-speech entry says those models leave “with at least three months’ notice.”
What steps should you follow to migrate?
These are suggested steps, not an OpenAI procedure.
- Inventory: search code, configuration files, scheduled jobs and notebooks for the exact model IDs in the first table, including any fine-tuned model trained from them. Also look for short aliases that resolve to these snapshots.
- Ask the API: call the List models or Retrieve a model endpoint and read shutdown_date for each ID in use.
- Pick a test set: choose 20 to 50 real requests covering structured outputs, tool calls and each language you serve.
- Compare before switching: run the same requests on the replacement. Compare accuracy, format validity, latency and cost. OpenAI sets the replacement’s price and behavior, so check its model page.
- Define “done” by real behavior: do not accept a migration only because tests pass. A published study shows how passing tests can hide a missing integration; see When Passing Tests Hide an Unfinished AI Coding Task.
- Set a date and a rollback: schedule the switch well before 14 January 2027. Keep a rollback path until the old ID stops responding.
Who is most at risk?
Teams with fine-tuned models, pinned snapshot IDs in scheduled jobs, or text-to-speech calls face the tightest deadlines. The text-to-speech models stop on 6 January 2027, eight days before the GPT-4 family (our arithmetic). If you only call a newer model, you can skip this article.
What to do next
- Open OpenAI’s deprecations page and find the entry dated 14 October 2026.
- List every place your product calls gpt-4o, gpt-4o mini, gpt-4.1 or gpt-4.1 mini, including fine-tuned models.
- Read the shutdown_date field for each model ID you use.
- Test gpt-6-luna on your own sample requests before you switch, and keep a rollback until the old ID stops responding.
Terms in plain language
- Model ID: the exact name your code sends to the API to choose a model.
- Snapshot: a model version tied to a date in its ID, such as 2024-08-06.
- Fine-tuned model: a model you trained further from a base model on your own examples.
- Endpoint: an API address your code calls to ask for something.
- Rollback: a way to return to the previous setup if the new one fails.
Questions and answers
When do the GPT-4o and GPT-4.1 API models shut down?
OpenAI lists 14 January 2027 for the model IDs named in its 14 October 2026 notice.
What does OpenAI recommend instead?
gpt-6-luna, for all four listed model families.
Are fine-tuned models affected?
Yes. Fine-tuned versions follow the same shutdown schedule as their base models.
Does this affect ChatGPT?
The cited entry concerns the API only. It does not describe changes to ChatGPT.
How can I check my own models?
Read the shutdown_date field through the List models or Retrieve a model endpoints.
Is the notice period shorter than usual?
Yes. The page says generally available models receive at least six months’ notice, while this entry gives about three months and does not explain why.
Sources
Facts last checked against OpenAI’s deprecations page on 15 October 2026 and re-checked on 29 October 2026. Check the page again before you plan, because dates and replacements can change.
Part of the guide: AI Model Pricing and API Cost Guides
