GPT-4o and GPT-4.1 retire on 14 January 2027: what to do before the date
OpenAI removes the gpt-4o, gpt-4o-mini, gpt-4.1 and gpt-4.1-mini snapshots from the API on 14 January 2027. The recommended replacement for all of them is gpt-6-luna. Many production support bots, RAG apps and agents still run on these, mostly the mini models. OpenAI does not say requests will be routed to the replacement: switch and re-test your own tasks before the date.
The models, by ID
| Model | Snapshot IDs | Shutdown | Recommended replacement |
|---|---|---|---|
| GPT-4ogpt-4o-2024-08-06, gpt-4o-2024-11-20 | gpt-4o-2024-08-06, gpt-4o-2024-11-20 | 14 Jan 2027 | gpt-6-luna |
| GPT-4o minigpt-4o-mini-2024-07-18 | gpt-4o-mini-2024-07-18 | 14 Jan 2027 | gpt-6-luna |
| GPT-4.1gpt-4.1-2025-04-14 | gpt-4.1-2025-04-14 | 14 Jan 2027 | gpt-6-luna |
| GPT-4.1 minigpt-4.1-mini-2025-04-14 | gpt-4.1-mini-2025-04-14 | 14 Jan 2027 | gpt-6-luna |
Read from OpenAI's API deprecations page on 12 November 2026 (notice dated 14 October 2026). Fine-tuned versions follow the same schedule as their base models. Aliases: the table lists dated snapshots only, but OpenAI's model pages showed the aliases gpt-4o, gpt-4o-mini and gpt-4.1-mini pointing to these deprecated snapshots, and OpenAI does not say they will be moved. If you call a bare alias, treat it as affected. gpt-4.1-nano already shut down on 23 October 2026. All dates: model retirement calendar.
Azure OpenAI has its own schedule. On 12 November 2026 Microsoft's Foundry model retirement schedule (last updated 23 September 2026) listed 14 April 2027 for gpt-4o 2024-08-06 and 2024-11-20, gpt-4o-mini 2024-07-18, gpt-4.1 and gpt-4.1-mini 2025-04-14, with gpt-5.1 as the replacement for gpt-4o and none named for the others. Fine-tuned deployments run to October 2027.
What happens if you do nothing
- On OpenAI's API, the model stops answering. OpenAI's page says: “At the time of the shut down, the model or endpoint will no longer be accessible.” It does not say requests are routed to gpt-6-luna. Plan for failed calls, not a silent switch.
- Fine-tunes go too. They shut down with their base model, and OpenAI says even active customers cannot create new fine-tuning jobs from 6 January 2027.
- On Azure, it depends on the deployment. Microsoft's lifecycle page says retired models return “410 Gone”. Standard-type deployments are auto-upgraded unless set to NoAutoUpgrade, and provisioned ones are not; but Microsoft lists a replacement only for gpt-4o (gpt-5.1), none for gpt-4o-mini or the gpt-4.1 family, so plan that move yourself. Either way, test first.
- Defaults. Frameworks and no-code tools often send gpt-4o-mini unasked. Check what each sends.
What happens to cost per finished task
OpenAI's model pages list gpt-6-luna at $0.10 input and $0.50 output per million tokens, against $0.15 / $0.60 for gpt-4o-mini and $0.40 / $1.60 for gpt-4.1-mini (12 November 2026). Cheaper per token is not cheaper per task: gpt-6-luna reasons by default (reasoning.effort medium), so answers may use more tokens and take longer, and a wrong answer that needs a retry or a person costs more than the tokens. Measure cost per finished task at the setting you will run.
Migration checklist
- Find every call site and model ID. Search code, config, environment variables and framework defaults for gpt-4o, gpt-4.1, aliases, dated snapshots and fine-tunes (ft: IDs).
- Freeze 20–30 real tasks with expected answers, including the cases that must never go wrong.
- Run old and candidate setups on the same tasks: your current model and gpt-6-luna at two reasoning settings, or another alternative. Same prompts, tools and grading.
- Compare cost per finished task and the named tasks that change. Count retries and failures in the cost; list the tasks each setup gets wrong.
- Check the mechanics: tool calls, JSON and structured output, refusals, latency, rate limits, token use.
- Roll out behind a flag. Move a share of traffic, then switch fully before 14 January (14 April on Azure).
- Keep the evidence: task set, results and decision, for the next retirement.
Why averages mislead
In our public test of the Aider coding agent on 27 tasks with hidden tests, Claude Sonnet solved the same 26 tasks as Claude Opus at 49% lower cost per solved task. Claude Haiku was far cheaper but failed 4 more tasks, mostly edge cases. An average hides which 4.
That test used Claude models, not OpenAI. The method is the same for a move from gpt-4o-mini or gpt-4.1-mini to gpt-6-luna: same tasks, known answers, cost per finished task, failures by name. Full method and raw data: Aider cost vs quality report.
Test your switch before 14 January
Free pilot: send 20–30 of your tasks with expected answers. We run 3 setups (your current GPT-4o or GPT-4.1 setup and two candidates) and send a one-page report: cost per finished task and every task each setup gets wrong.
7-day test, $1,390 fixed: 20 tasks, 3 setups, a signed report in 7 working days with a recommendation.
24/7 watch, $169 or $349 a month: after the move we re-run your frozen tasks and e-mail you when results or cost per task change, including silent model changes.
We work in writing: use the “Write to us” chat on this page or e-mail hello@agenttrustapi.com.
For agents: the calendar is also a JSON file and an Atom feed, each with source and check date. Related guides: GPT-5 and o3 retire on 11 December 2026 · Gemini 2.5 retires on 1 December 2026.
Questions people ask
When are GPT-4o and GPT-4.1 retired on the OpenAI API?
On 14 January 2027: 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, per OpenAI's API deprecations page (checked 12 November 2026).
I call plain gpt-4o-mini or gpt-4.1-mini. Am I affected?
Treat yourself as affected. OpenAI's model pages show these aliases pointing to the deprecated snapshots, and OpenAI does not say they will be moved to a new model.
What replaces GPT-4o and GPT-4.1?
OpenAI recommends gpt-6-luna for all five snapshots. It is cheaper per token than gpt-4o-mini but reasons by default, so measure cost per finished task and test one alternative too.
Will my requests fail after 14 January?
Plan for that. OpenAI says that at shutdown the model will no longer be accessible and does not say requests are routed to a replacement. Fine-tunes follow their base model.
Are the Azure OpenAI dates the same?
No. Microsoft's Foundry schedule listed 14 April 2027 for these versions (checked 12 November 2026). Retired Azure models return 410 Gone; Standard deployments auto-upgrade where Microsoft lists a replacement (gpt-4o: gpt-5.1); for gpt-4o-mini and the gpt-4.1 family it lists none, so plan the move yourself.
How do I test the switch before the deadline?
Freeze 20–30 real tasks with expected answers, run your current setup and the candidates, compare cost per finished task and the named tasks that change, and roll out behind a flag. We do this as a free pilot or a 7-day test for $1,390.
Questions or corrections: hello@agenttrustapi.com.