← Back to Insights

OpenAI is shutting off older AI models for apps, some this Friday and GPT-4o on Jan 14. Here's how to check your business.

On Oct 14, OpenAI gave developers three months' notice that GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini leave its API on Jan 14, 2027. If a chatbot, workflow or plugin in your business was built in the last two years, it may be running on one of them. Here's how to find out and switch safely before the holidays.

OpenAIGPT-4oGPT-4.1GPT-6 LunaAI automationmodel migrationoperations

What happened

On October 14, 2026, OpenAI notified developers that four GPT-4 series models will be removed from its API on January 14, 2027: GPT-4o (gpt-4o-2024-08-06, gpt-4o-2024-11-20), GPT-4o mini (gpt-4o-mini-2024-07-18), GPT-4.1 (gpt-4.1-2025-04-14) and GPT-4.1 mini (gpt-4.1-mini-2025-04-14). Fine-tuned versions of those models shut down the same day. OpenAI lists gpt-6-luna as the recommended replacement for all four.

Three details matter if you run the business:

  • It's the API, not the ChatGPT app. ChatGPT users already lost these models on Feb 13, 2026. This notice hits the things that call OpenAI in the background: website chatbots, email and document workflows, no-code automations, and tools someone built for you.
  • The runway is short. OpenAI's own policy promises at least six months' notice for generally available models. The New Stack points out this one gave about three, and the window runs through the December holidays.
  • There's an earlier deadline this Friday. Under an April notice, older entries including gpt-4o-2024-05-13, gpt-4.1-nano, gpt-4-turbo, gpt-3.5-turbo-0125, o1, o3-mini, o4-mini and gpt-image-1 shut down on October 23, 2026.

The replacement is cheaper on paper. OpenAI lists GPT-6 Luna at $0.10 per million input tokens and $0.50 per million output tokens, compared with $2.50 / $10 for GPT-4o on OpenAI's own GPT-4o model page. But it isn't a drop-in swap. Luna reasons by default (medium effort), it supports function calling in the older Chat Completions API only with reasoning set to "none", it doesn't support fine-tuning, and prompts over 272K input tokens are billed at higher rates. Microsoft's Azure OpenAI runs its own schedule; DataNorth reports that Azure lists April 14, 2027 for the same models.

Who it helps

  • Owners who had an AI feature built in 2024 or 2025 (a site chatbot, a lead-sorting flow, an invoice or document reader) and aren't sure what's under the hood.
  • Teams running Zapier, Make, n8n or Power Automate flows with an OpenAI step, where the model name sits in a setting nobody has opened since setup.
  • Anyone with a fine-tuned GPT-4o or GPT-4.1 model. This group has the most work to do, because there's no fine-tunable replacement. OpenAI's self-serve fine-tuning stops taking new jobs for active customers on Jan 6, 2027.
  • Tech-services firms and freelancers who built AI tools for clients and now need to find and fix them before they break.

How to apply

  1. List what depends on AI (10 minutes). Write down every process where AI does part of the work: replies drafted, leads tagged, documents turned into fields, weekly reports. Include tools a vendor or freelancer set up.
  2. Find the model names (10 minutes). Search your code, automation steps and plugin settings for gpt-4o, gpt-4.1 and o4-mini. If you have an OpenAI API account, OpenAI's List models endpoint now shows a shutdown_date for each model, and the deprecations page has the full table. For vendor tools, email the vendor and ask which model they use and when they'll switch.
  3. Handle Friday's deadline first. Anything on gpt-4o-2024-05-13, gpt-4.1-nano or the other Oct 23 entries needs attention this week.
  4. Test before you switch. Save 10 to 20 real inputs: a normal request, a messy one, and one that should be rejected. Run them through gpt-6-luna with reasoning set to "none" first, and compare the output and format with what you get today. A correct answer in the wrong field can still break the next step.
  5. Switch with a way back. Change one process at a time, keep the old setting written down until Jan 14, and check results daily for the first week. Track cost per completed task, since reasoning tokens count toward your bill if you raise the effort.
  6. Don't plan around an extension. OpenAI mentions dedicated capacity through its sales team, but most owners should aim to be done well before the holidays.

How CerpaMedia can help

If you'd rather not dig through settings and code yourself, we can do the whole switch for you. We find every place a retiring model is set, including code, no-code flows, website plugins and vendor tools. Then we test the replacement side by side on your real examples, move each process over with a rollback ready, and hand you a one-page record of what runs where and on which model. We also add a simple shutdown-date check, so the next retirement notice shows up as a calendar item instead of an outage. Most of the risk here is in finding the dependencies, so that's where we start.

Sources