Skip to main content
All articles
Practical GuideOctober 23, 2026·4 min read

Is Your AI Automation About to Break? The Fall 2026 Model Retirement Checklist

OpenAI shuts off GPT-4, o1 and other older models on October 23, with Claude Sonnet 4.5 and GPT-4o next. A plain-English checklist to keep automations running.

Most software breaks when someone changes it. AI automations can break when nobody touches them at all. Today, October 23, OpenAI switches off access to GPT-4, GPT-3.5 Turbo, o1 and a list of other older models, a shutdown it announced six months ago. Two more deadlines follow: Claude Sonnet 4.5 on November 30, and GPT-4o and GPT-4.1 on January 14. If a Zapier, Make or n8n flow, a chatbot or a plugin you rely on calls one of them by name, it stops working on that date.

Why Working Automations Break

Your tools reach an AI model through an API, the connection software uses to send work to the model and get answers back. Every request names one exact model, such as "gpt-4" or "claude-sonnet-4-5-20250929." When the provider retires that model, the request has nowhere to go. Anthropic's documentation puts it plainly: "Requests to retired models fail." Nothing warns your customers first. The chatbot stops answering, or the automation step fails.

The Three Dates to Know

  • October 23, 2026 (today), OpenAI: gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4.1-nano, the May 2024 version of GPT-4o (gpt-4o-2024-05-13), gpt-image-1, o1, o1-pro, o3-mini and o4-mini, plus fine-tuned versions of several of them. (A fine-tuned model is a copy trained further on your own examples.) For the text models, OpenAI suggests GPT-5.6 Sol, Terra or Luna as substitutes.
  • November 30, 2026, Anthropic: Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) retires from the Claude API. Anthropic notified developers on September 30 and recommends Claude Sonnet 5.5.
  • January 14, 2027, OpenAI: GPT-4o (the August and November 2024 versions), GPT-4o mini, GPT-4.1 and GPT-4.1 mini leave the API. OpenAI announced this on October 14 and recommends GPT-6 Luna. Fine-tuned versions of these models follow the same schedule.

The Quieter Break: Routine Updates

Two days ago, Anthropic released a new API version, a dated set of rules for how requests must be written. New versions of its developer toolkits (SDKs, the code libraries developers use to talk to Claude) use it by default and bring breaking changes of their own. Under the new version, a request that includes temperature, top_p or top_k, settings that control how varied the replies are, returns an error "even at the parameter's default value."

The older version still works, so nothing breaks on its own. The risk is the routine "update everything" a developer runs on a quiet afternoon. If you have a custom Claude chatbot or app, ask whoever maintains it to test new toolkit versions on a copy before updating the live one.

Where Do You Stand?

Does anything you run call OpenAI or Claude with your own API key? Think custom apps, AI steps in automation tools, chatbots you set up, plugins.
No. Your software vendors carry this. Ask each one, in writing, which models they use and how they are handling these dates.
Yes. Continue.
Does any of them name a model from the three dates above?
No. You are clear for now. Check again each quarter.
Yes. Switch to the recommended replacement and test it on real examples before the date. Anything on today's list is due now.

Your own API key means your own deadline. A vendor's key means the vendor's.

A 30-Minute Checklist

  1. List every tool that uses AI: automations, chatbots, plugins and anything a freelancer built for you.
  2. Open each AI step and write down the model name. If you cannot find it, ask whoever built it.
  3. Match the names against the three dates. Anything on today's list needs attention now.
  4. Switch, then test five real examples. Replacements can behave differently; Anthropic's own release notes list breaking changes when moving to its newer models.
  5. Hold toolkit updates on custom Claude apps until someone tests them.
  6. Set a quarterly reminder to read OpenAI's and Anthropic's deprecation pages, where both list upcoming shutdowns.

When we build workflow automation for clients, the model name lives in one setting with a named owner, so a retirement notice becomes a five-minute change instead of an outage.

Sources

  1. OpenAI, "Deprecations" (October 14, 2026): the October 23 shutdown list announced April 22, the January 14, 2027 removals, replacements and fine-tuned models. developers.openai.com
  2. Internet Archive, snapshot of OpenAI's "Deprecations" page (October 16, 2026): the scheduled October 23 shutdown and the October 14 notice, as published before today. web.archive.org
  3. Anthropic, "Model deprecations" (September 30, 2026): Claude Sonnet 4.5 retirement on November 30, its replacement, and that requests to retired models fail. platform.claude.com
  4. Anthropic, "Claude Platform release notes" (October 21, 2026): the new API version and toolkit versions with breaking changes; earlier entries list breaking changes for newer models. platform.claude.com
  5. Anthropic, "Migrating to API version 2026-10-21" (undated guide linked from the October 21 release notes): removed settings, and keeping requests on the older version. platform.claude.com

Not sure which models your automations call?

We will inventory every AI step in your tools, flag anything on a retirement list, and move it to a supported model with a test, before a deadline turns into an outage.

Find Your Fit Book a Free Call