AI Model Retirement: What Happens to Your App When Its AI Model Is Switched Off

AI model retirement is the date a provider stops serving a model, and from that day any feature in your app that calls it either stops working or starts running on a model you didn't choose. Which of the two happens depends on how your app reaches the model, so the first job is finding out.

The next AI model retirement dates are close. Lovable retired Gemini 2.5 Pro, Flash and Flash Lite for the AI features in its apps on October 20, 2026, and its deprecated models list puts Gemini 3.6 Flash next, on November 19. Anthropic retires Claude Sonnet 4.5 on its API on November 30, and OpenAI removes GPT-5, GPT-5 mini and GPT-5 nano from its API on December 11.

What happens on an AI model retirement date?

One of two things: the request fails, or it is quietly sent to a replacement.

Hard failure. Anthropic's model deprecations page says requests to retired models fail, and OpenAI's deprecations page says a shut-down model is no longer accessible. A chatbot that calls one of those models directly returns errors until someone changes the code.

Silent replacement. For many retired models, Lovable sends your app's requests to a replacement, so the feature keeps working without a code change. Lovable warns that the replacement "can cost a different amount per request and respond differently." For others on its list, such as GPT-4o mini TTS on January 6, 2027, requests simply fail.

In our view, the silent version of AI model retirement is the riskier one. A broken feature gets noticed quickly. A support assistant that starts answering differently, or costs more per conversation, can run for weeks before anyone looks.

Which AI features in your app depend on a model?

Any feature that sends data to an AI provider and acts on the reply: a chat assistant, a summary step on uploaded documents, a classifier that routes leads, image or speech features. Each one is exposed when its model is retired.

List each feature, the exact model name it calls, and the route it takes, because each route has its own dates:

• Through an app builder. Lovable lists every deprecated model and its date on one page, and shows the model behind each request under More, then AI. It can also retire a model before the provider does.
• Direct to the provider. OpenAI's List models endpoint returns a shutdown_date for each model. Anthropic's Console exports usage by API key and model.
• Through a cloud provider. Anthropic notes that Amazon Bedrock and Google-managed Claude set their own retirement schedules.

If nobody can name the model behind a feature, that is your first finding. The same inventory habit helps with forgotten AI apps your team built and moved on from.

How much warning do you get?

Usually months, and the notice goes to whoever owns the account.

OpenAI commits to at least six months for generally available models, three for specialized variants, and as little as two weeks for preview models. Anthropic gives at least 60 days for publicly released models. Lovable emails the owner of every project whose app calls a model once it has a retirement date, and says that warning can be short.

So check whose inbox those emails reach. If a freelancer opened the API account, or the account owner has left, the AI model retirement notice lands with them and nobody at your company sees it.

What should you do before the AI model retirement date?

Switch on your schedule, not the provider's. A model change is a small release, and it deserves a small test.

1. Choose the replacement yourself. Start from the provider's recommendation: OpenAI lists GPT-5.6 Sol for GPT-5, and Anthropic lists Claude Sonnet 5.5 for Sonnet 4.5. Lovable advises picking a model rather than relying on its reroute.
2. Test with real examples. Save 15 to 20 real inputs per feature, with answers you were happy with, and compare the new model's output on format, accuracy and tone.
3. Check the cost per request. Run the same test set, compare usage with last month, and set a spending alert.
4. Keep the model name in one place. A single configuration setting turns the next switch into a one-line change instead of a hunt through the code.
5. Keep preview models away from customers. OpenAI advises against them for business-critical work unless you can migrate on short notice.
6. Put the dates on a calendar, with a reminder a month before each.

Frequently asked questions

Is a deprecated model the same as a retired one?

No. A deprecated model still works but has an end date. A retired model no longer answers, so requests fail or get rerouted.

Will my app break if I do nothing?

It depends on the route. Direct calls to OpenAI or Anthropic fail after the shutdown date. Lovable reroutes many models, so the feature keeps running on a model you didn't pick.

Can I keep using a retired model?

Usually not once the AI model retirement date passes. OpenAI says some customers may arrange dedicated capacity through its sales team, which won't suit most small businesses.

How often does AI model retirement happen?

Several times a year. OpenAI's page alone lists shutdowns in July, October and December 2026, and more in early 2027. Check your list each quarter.

Who should own model updates in a small company?

Whoever owns the app's other maintenance. AI model retirement belongs on the same list as security patches and dependency updates.

Treat model changes as routine maintenance

AI model retirement is now a normal part of running an app with AI features, like renewing a certificate. List your features and their models this week, and put the next date in the calendar. If your app was built with an AI tool and nobody is sure what it calls, our Prototype to Production team can map it, or you can start with a free audit.