GPT-4o API Retirement: What Businesses Must Do Before January 2027

OpenAI has announced that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from its API on 14 January 2027. The notice was published on 14 October 2026 on OpenAI's official deprecations page, and the recommended replacement for all four is a newer model called gpt-6-luna.

If your website chatbot, lead qualifier, support assistant, content tool or internal automation was built in the last two years, there is a good chance it calls one of these models. After the shutdown date, those calls will stop working. This guide explains what is changing, how to find out if you are affected, and a simple plan to migrate without breaking anything your customers rely on.

What exactly is being retired

According to OpenAI's deprecations page, the "GPT-4 series" entry dated 14 October 2026 covers these model snapshots, all with removal from the API on 14 January 2027:

OpenAI also notes that fine-tuned versions follow the shutdown schedule of their base models. So if you paid to fine-tune GPT-4o or GPT-4o mini on your own data, that custom model is on the same clock.

Two related changes are worth knowing about because they often appear in the same projects:

• GPT-4.1 nano was covered by an earlier notice (22 April 2026) with a shutdown date of 23 October 2026, and gpt-5.6-luna listed as the replacement. If you used it, it should already have been replaced.
• The Assistants API was scheduled for removal on 26 August 2026, with the Responses API and Conversations API as replacements. Tools built on the old Assistants API needed a rebuild of their conversation handling, not just a model swap.

This change is about the API, which is what developers use to build AI into websites and software. It is separate from which models you see inside the ChatGPT app.

Who is affected

You are likely affected if any of these are true:

• Your website has an AI chat widget or assistant that was custom built, or built by a freelancer or agency, using OpenAI.
• You use a no-code automation (for example a workflow that drafts emails, summarises leads or writes product descriptions) where someone selected "gpt-4o" or "gpt-4o-mini" as the model.
• Your CRM, help desk or WhatsApp bot was connected to OpenAI with your own API key.
• Your team built internal tools, such as a document Q&A tool, on GPT-4o or GPT-4.1.

You are probably not directly affected if you only use off-the-shelf software where the vendor manages the AI for you. Even so, it is worth sending your vendors a one-line question: "Do any of your features rely on GPT-4o or GPT-4.1, and what is your plan before 14 January 2027?"

How to check in 15 minutes

You do not need to be technical to run a first check.

1. Look at your OpenAI usage dashboard. If you have an OpenAI account with an API key, the usage section shows which models your keys are calling. Note every model name that appears in the table above.
2. Search your code or settings for model names. Ask your developer to search the codebase for gpt-4o, gpt-4o-mini and gpt-4.1. In no-code tools, open each AI step and check the "model" field.
3. Check short aliases too. Many projects use a short name such as gpt-4o instead of a dated snapshot. Confirm with OpenAI's documentation or your developer which snapshot an alias points to, and treat it as affected until confirmed otherwise.
4. List what each AI feature does for the business. Lead capture, customer support and order updates matter more than an internal summary tool. This list becomes your migration priority order.

A safe migration plan

Swapping one model name for another sounds like a two-minute job, and sometimes it is. But newer models can respond in a different tone, length or format, and that can quietly break things like a form that expects a specific answer structure. Here is the approach we recommend.

1. Freeze a test set first

Before changing anything, collect 20 to 50 real examples of what your AI feature handles: typical customer questions, awkward ones, and the ones where a wrong answer would be costly (pricing, refunds, medical or legal topics). Save the current answers. This is your baseline.

2. Test the replacement side by side

Run the same examples through the replacement model OpenAI recommends, gpt-6-luna, and compare. Check for:

• Accuracy against your own information (prices, opening hours, policies)
• Tone of voice and answer length
• Structured outputs, such as JSON your website or CRM reads
• Refusals or over-cautious answers on normal business questions
• Speed, which matters for live chat

3. Check the cost

Pricing differs between models, and a newer model is not always cheaper per request. Look up the current price on OpenAI's pricing page, multiply by your monthly volume, and set a spending limit in your OpenAI account before switching production traffic.

4. Update prompts, then switch gradually

Most migrations need small prompt edits to restore the tone and format you had. Once the test set looks right, move a small share of real traffic to the new model, watch the results for a few days, then switch fully. Keep the old configuration documented so you can explain what changed.

5. Do not wait until January

Do not leave this until the last week. Holiday season traffic, code freezes and developer leave all land in December. Aim to finish testing in November and complete the switch in early December.

Use this as a chance to improve, not just patch

A forced migration is a natural moment to ask whether your AI setup still fits the business. Two questions are worth asking.

Is a simple chatbot still enough? Many 2024 era chatbots only answer questions. Today, businesses increasingly use AI agents that can also take actions, such as booking an appointment, qualifying a lead into your CRM or sending a WhatsApp follow-up, with a human approving anything sensitive. Our guide to AI agents for business explains the difference.

Are you locked into one provider? The model landscape changes every few months. If your system is built so the model is a setting rather than hard-wired throughout the code, the next retirement notice becomes a routine update. That design choice is one of the cheapest forms of insurance you can buy.

Keep an eye on the dates of your other tools too. OpenAI's deprecations page lists retirements across its products, and other AI providers publish similar schedules. Put a quarterly reminder in the calendar to check them.

Quick checklist

• Identify every feature that calls GPT-4o, GPT-4o mini, GPT-4.1 or GPT-4.1 mini
• Ask software vendors about their migration plans
• Build a test set of real questions and save current answers
• Test gpt-6-luna side by side and adjust prompts
• Confirm pricing and set spending limits
• Switch gradually and monitor
• Finish well before 14 January 2027

Frequently asked questions

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

OpenAI's deprecations page lists removal of GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini from the API on 14 January 2027. Requests to those models will fail after that date.

What should I replace GPT-4o with?

OpenAI lists gpt-6-luna as the recommended replacement for all four models. Test it on your own real examples before switching, because tone, format and cost can differ.

Will my fine-tuned GPT-4o model keep working?

No. OpenAI states that fine-tuned models follow the shutdown schedule of their base models, so a model fine-tuned on GPT-4o or GPT-4o mini is also affected.

Does this affect ChatGPT for my team?

This notice is about the developer API used to build AI into websites and software. Check OpenAI's own announcements for changes to the ChatGPT app separately.

The bottom line

If your business uses AI built on OpenAI, check this month whether any of it runs on GPT-4o or GPT-4.1, and plan the move before 14 January 2027. Done carefully, the migration takes days, not weeks, and it is a good opportunity to turn a basic chatbot into a more useful AI agent. If you would like us to audit your setup or handle the migration, see our AI agents service or get in touch.