OpenAI Will Shut Off GPT-4o and GPT-4.1 in the API on January 14, 2027, Giving Developers 3 Months When Its Policy Promises 6

OpenAI told developers on October 14, 2026 that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from its API on January 14, 2027. Fine-tuned versions of those models go dark on the same day. The notice sits at the top of OpenAI’s API deprecations page, and the recommended replacement for all four is a single model: gpt-6-luna.

That gives teams 92 days. The same OpenAI page says generally available models get “at least 6 months” of notice unless safety or compliance concerns require a faster timeline. The October 14 entry gives no reason for the shorter window.

For most apps the move will lower the bill on paper. For anyone running a fine-tuned GPT-4o or GPT-4.1, or leaning on non-reasoning speed, it is a real migration project with a hard date.

Exactly what OpenAI is shutting down

The deprecation table lists dated snapshots, not bare model names. Here is the full list from OpenAI’s notice.

Model	Snapshot IDs being removed	Shutdown date	Recommended replacement
GPT-4o	gpt-4o-2024-08-06, gpt-4o-2024-11-20	January 14, 2027	gpt-6-luna
GPT-4o mini	gpt-4o-mini-2024-07-18	January 14, 2027	gpt-6-luna
GPT-4.1	gpt-4.1-2025-04-14	January 14, 2027	gpt-6-luna
GPT-4.1 mini	gpt-4.1-mini-2025-04-14	January 14, 2027	gpt-6-luna

One detail matters if your code just says gpt-4o. OpenAI’s GPT-4o model page says that alias points to gpt-4o-2024-08-06, one of the snapshots on the list. The GPT-4o mini, GPT-4.1 and GPT-4.1 mini aliases also point to listed snapshots. The notice does not spell out what the bare names will do after January 14, so the safe plan is to treat them as going away too.

The oldest GPT-4o snapshot, gpt-4o-2024-05-13, is not part of this notice because it already has an earlier date. It shuts down on October 23, 2026, nine days from now, along with gpt-4-turbo, gpt-4.1-nano, o3-mini and o4-mini.

Is GPT-6 Luna a fair swap? The price math

On list price, yes. OpenAI’s pricing page puts every retiring model above Luna on both input and output.

Model (Standard tier)	Input / 1M tokens	Cached input / 1M	Output / 1M
GPT-4o	$2.50	$1.25	$10.00
GPT-4.1	$2.00	$0.50	$8.00
GPT-4.1 mini	$0.40	$0.10	$1.60
GPT-4o mini	$0.15	$0.075	$0.60
GPT-6 Luna	$0.10	$0.01	$0.50

Take a support bot that reads 50 million tokens and writes 10 million a month. On GPT-4o that is about $225. On GPT-4.1 it is about $180. On Luna, with the same output, it is about $10.

The catch is that Luna is a reasoning model and the old ones are not. The Luna model page lists reasoning effort settings from none to max, with medium as the default. OpenAI’s reasoning guide says reasoning tokens are billed as output tokens.

That matters most for GPT-4o mini users. The same 50 million in, 10 million out workload costs about $13.50 on GPT-4o mini. If Luna’s default reasoning adds another 10 million output tokens, the Luna bill rises to about $15, slightly more than what you pay today. Set the effort level on purpose instead of taking the default.

Two more pricing rules to check. Luna bills cache writes at 1.25 times the input rate, and prompts over 272,000 input tokens cost 2 times on input and 1.5 times on output for the whole request. GPT-4.1 users who feed in very long documents should rerun their numbers.

Migration gotchas buried in the docs

Swapping the model string is not the whole job. A few differences show up when you put the GPT-4o and Luna model pages side by side.

• Function calling in Chat Completions. OpenAI says Luna supports function calling on Chat Completions only when reasoning_effort is set to none. Built-in tools need the Responses API.
• No fine-tuning. GPT-4o lists fine-tuning as supported. Luna’s endpoint table marks fine-tuning as not supported.
• Predicted outputs. GPT-4o lists predicted outputs as a feature. Luna’s feature list does not include it.
• Bigger limits. Luna offers a 1,050,000 token context window and 128,000 max output tokens, against 128,000 and 16,384 on GPT-4o.
• Newer knowledge. Luna’s cutoff is May 18, 2026. GPT-4o’s is October 1, 2023 and GPT-4.1’s is June 1, 2024.

OpenAI also gives teams a way to catch this in code. The deprecations page says GET /v1/models now returns an optional shutdown_date field for each model, so a nightly check can flag any model ID with a date attached.

Fine-tuned GPT-4o users get the hardest deal

This is the part of the notice that will hurt. Fine-tuned versions of the four models “follow the same shutdown schedule as their underlying base models,” the deprecations page says. So a custom GPT-4o or GPT-4.1 model built over months stops answering on January 14.

There is no like-for-like path inside OpenAI. Luna cannot be fine-tuned, and OpenAI is already winding down self-serve fine-tuning: active customers lose the ability to create new jobs on January 6, 2027. Teams will have to rebuild that behaviour with prompts, examples and evaluation sets, or move the workload elsewhere.

Microsoft offers some breathing room. Its Foundry retirement schedule lists April 14, 2027 for the same GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini versions, and keeps fine-tuned GPT-4o deployments running until October 1, 2027. That page was last updated on September 21, before OpenAI’s notice, and most of those versions are already closed to new Azure customers.

Confirmed vs unconfirmed

Claim	Status	Source
Four model families leave the OpenAI API on January 14, 2027	Confirmed	OpenAI deprecations page
Developers using them were notified on October 14, 2026	Confirmed	OpenAI deprecations page
Fine-tuned versions shut down on the same date	Confirmed	OpenAI deprecations page
gpt-6-luna is the recommended replacement for all four	Confirmed	OpenAI deprecations page
Luna is cheaper per token than all four	Confirmed on list price; real cost depends on reasoning effort	OpenAI pricing and reasoning guide
Why notice is 3 months, not the 6 the policy states	Not explained	No reason given in the notice
What bare aliases like gpt-4o do after January 14	Not stated; aliases point to retiring snapshots	OpenAI model pages
Azure keeps these versions until April 14, 2027	Listed as of September 21; may change	Microsoft Learn

Why people are upset beyond the code

GPT-4o already left ChatGPT. OpenAI retired it from ChatGPT on February 13, 2026, saying only 0.1% of users still picked it each day. The API was the last official route to the model.

Within hours of the notice, the #keep4o campaign was back on X. The Relational AI Alliance wrote that “the last official path to GPT-4o has a shutdown date.” For developers the issue is more practical, but the sentiment explains why this deprecation is trending when most are ignored.

What it means for Indian developers

If your product runs on GPT-4o mini as a budget default, as many early chatbots and classifiers do, run a test pass on Luna with reasoning set to none or low first. Compare the token counts in your logs before switching all traffic, since the saving is small at this price level.

Agencies that sold fine-tuned GPT-4o or GPT-4.1 models to clients should tell those clients now. Three months is tight for rebuilding a custom model, testing it and getting sign-off, especially with the December holiday slowdown in between.

If you are weighing options outside OpenAI, Anthropic’s Claude Haiku 5.5 launched at the same $0.10 and $0.50 per million tokens, as we covered in our Claude Haiku 5.5 pricing breakdown. For the OpenAI side, our look at how GPT-6 Sol and Luna cut API prices explains where Luna sits in the lineup.

OpenAI’s deprecation calendar is getting crowded

This is the third API retirement notice from OpenAI in two weeks, and it lands on top of a long list of dates.

Date	What shuts down
October 23, 2026	gpt-4o-2024-05-13, gpt-4-turbo, gpt-4.1-nano, o1, o3-mini, o4-mini, gpt-image-1 and more
November 30, 2026	Agent Builder, the Evals platform and reusable prompt objects
December 11, 2026	Original GPT-5, GPT-5 mini, GPT-5 nano and o3 snapshots
January 6, 2027	tts-1, tts-1-hd and older TTS models; new fine-tuning jobs end
January 14, 2027	GPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini and their fine-tunes
January 20, 2027	Legacy audio and realtime models
February 26, 2027	whisper-1 and GPT-4o transcription models

We saw the same pattern in September, when OpenAI shut down four legacy models hours before DevDay. The direction is clear: OpenAI wants all API traffic on the GPT-6 family, and it is cleaning house fast. Luna also powers newer tools like the Decisions API, which tells you where OpenAI is putting its effort.

The practical advice is boring but important. Search your codebase and config for every gpt-4o and gpt-4.1 string, run your evaluation set on Luna this month, and do not wait until January.

FAQ

When does GPT-4o stop working in the OpenAI API?

January 14, 2027. OpenAI says the gpt-4o-2024-08-06 and gpt-4o-2024-11-20 snapshots will be removed that day. The older gpt-4o-2024-05-13 snapshot goes earlier, on October 23, 2026.

Which models are affected by the January 14, 2027 shutdown?

GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini, plus any fine-tuned models built on them. GPT-4.1 nano is on a separate October 23, 2026 date.

What should I replace GPT-4o with?

OpenAI recommends gpt-6-luna, priced at $0.10 per million input tokens and $0.50 per million output tokens. It is a reasoning model, so set the reasoning effort deliberately and test on your own data.

Can I keep my fine-tuned GPT-4o model?

Not on OpenAI’s API after January 14, 2027. Luna does not support fine-tuning. On Azure, Microsoft currently lists fine-tuned GPT-4o deployments until October 1, 2027.

Is GPT-4o still available in ChatGPT?

No. OpenAI removed GPT-4o from ChatGPT on February 13, 2026. The API was the last official way to use it.