OpenAI Is Retiring GPT-4o From the API on January 14, 2027. Developers Got 3 Months' Notice, Not the 6 It Promised
This is the second shutdown wave in three months. GPT-3.5 and the original GPT-4 line go dark on October 23, 2026. GPT-4o, the model still running inside countless production apps and Custom GPTs, gets its own cutoff barely 12 weeks later.
GPT-4o isn't some forgotten snapshot. OpenAI brought it back from a planned retirement in 2025 after a ChatGPT user backlash, with Sam Altman promising at the time, 'If we ever do deprecate it, we will give plenty of notice,' per The Register. Three months is the notice this time.
What's actually retiring
| Model | API access ends | Fine-tune training ends | Recommended replacement |
|---|---|---|---|
| gpt-4o-2024-08-06 / gpt-4o-2024-11-20 | Jan 14, 2027 | Jan 6, 2027 | gpt-6-luna |
| gpt-4o-mini-2024-07-18 | Jan 14, 2027 | Jan 6, 2027 | gpt-6-luna |
| gpt-4.1-2025-04-14 | Jan 14, 2027 | Jan 6, 2027 | gpt-6-luna |
| gpt-4.1-mini-2025-04-14 | Jan 14, 2027 | Jan 6, 2027 | gpt-6-luna |
All four map to gpt-6-luna, OpenAI's current low-cost model, priced around $0.10 per million input tokens and $0.50 per million output tokens on the standard tier, per pricing trackers citing OpenAI's own rate card. It isn't a drop-in swap: gpt-6-luna reasons and calls tools differently than GPT-4o did, so changing the model ID is the easy part of the migration.
Who actually has to move, and fast
- ✦Anyone with a production app hardcoded to gpt-4o or gpt-4.1. There's no automatic fallback. The call simply fails after January 14.
- ✦Fine-tunes built on either base model. Retraining against a new base has to start before January 6, when OpenAI stops accepting new fine-tuning jobs against the retiring models.
- ✦Teams that genuinely need more than three months. OpenAI's enterprise sales team can reportedly arrange dedicated capacity to keep a retiring model running past its public cutoff, per The New Stack, but that's a negotiated exception, not a default.
- ✦Anyone who remembers the 2025 GPT-4o petition. The six-month notice policy exists partly because of that backlash. This cutoff runs on half of it.
Before January 14
- 1Search your codebase for the exact snapshot IDs: 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.
- 2Test gpt-6-luna in staging before touching production. Its tool-calling and reasoning behavior differs enough from GPT-4o that a straight model-ID swap can change your output.
- 3Start any required retraining before January 6, 2027. That's when new fine-tuning submissions against the old base models stop being accepted, not when the models themselves retire.
- 4Call OpenAI's sales team now if three months genuinely isn't enough runway. Dedicated capacity is reportedly negotiable in December; it won't be in January.
When does GPT-4o actually leave the OpenAI API?
January 14, 2027, per OpenAI's own deprecations page. That covers 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, plus fine-tunes built on any of them.
Does this affect ChatGPT, or just the API?
This notice is API-only. ChatGPT itself already removed GPT-4o from its model picker in February 2026, with roughly two weeks' notice, per The Register's reporting at the time.
What should I replace GPT-4o or GPT-4.1 with?
OpenAI's own documentation points to gpt-6-luna, its current low-cost model. Test it in staging first: it handles reasoning and tool calls differently than GPT-4o, so output can change even with an identical prompt.
Why is the notice only three months instead of six?
OpenAI's deprecation policy promises at least six months for a generally available model, with exceptions for safety or compliance reasons. OpenAI hasn't publicly explained why GPT-4o and GPT-4.1 are getting roughly half that window.
What happens to a fine-tuned model built on GPT-4o?
It keeps serving inference requests until January 14, 2027, same as the base model. New fine-tuning jobs against gpt-4o or gpt-4.1 stop being accepted on January 6, 2027, so retraining against a different base has to start before then.
Try the tools ✦
Free browser tools that never upload your files.