Your AI model is retiring: how to keep your automations from stalling on 1 December
On 30 November, OpenAI shuts down Agent Builder, Evals and reusable prompts, and Claude Sonnet 4.5 stops in the API. What does that mean for your chatbot, flows and scripts, and what should you arrange now?
AI models are retiring. And it's happening sooner than many entrepreneurs think. Your chatbot has been running fine for months, the n8n flow faithfully processes every request, the script neatly summarises your emails. Then one morning nothing but error messages comes through. Not because anything is broken, but because the model underneath no longer exists.
On 30 November 2026, several of these moments coincide. If you use AI via OpenAI or Anthropic, it's wise to take a look under the hood this month. Below you can read what is being switched off, what already went wrong earlier, and how to make sure your processes keep running smoothly on 1 December.
What disappears on 30 November
At OpenAI it isn't about one model, but about a piece of tooling. On 3 June 2026, OpenAI announced that Agent Builder and Evals are being phased out. From 30 November they will no longer be available on the platform. Evals goes in two steps: existing evals have been read-only since 31 October, and on 30 November the dashboard and the API stop. On the same day, the v1/prompts API and the reusable prompt objects also disappear.
Good to know: the underlying Responses API simply continues to exist. What disappears is the Builder interface with the agents you saved in it. As alternatives, OpenAI points to the Agents SDK for workflows in code, to Workspace Agents in ChatGPT for use cases that work better with plain-language prompts, and to Promptfoo as a replacement for Evals. What is still unclear: OpenAI hasn't said how long the history of your eval runs will be kept after 30 November. So export it.
At Anthropic it really is about a model. On 30 September 2026, the company announced that Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) will be removed from the Claude API on 30 November. The advice is to switch to Claude Sonnet 5.5. After that date, every request that still calls that old ID will fail. That date applies to the Claude API, Claude Platform on AWS and Microsoft Foundry. Amazon Bedrock and Google Cloud follow their own schedule.
23 October already showed what happens
A preview came earlier this autumn. On 23 October, OpenAI switched off a series of older models in its API, including o1, o1-pro, o3-mini, o4-mini, GPT-4 Turbo, the original GPT-4, gpt-4o-2024-05-13, gpt-4.1-nano and gpt-3.5-turbo-0125. gpt-image-1 and old fine-tunes based on those models were also affected. This had been announced on 22 April: more than half a year of warning.
The sources don't entirely agree on the exact number of models or the recommended replacements, by the way. So always use OpenAI's official deprecations page as your guide. And it doesn't stop there: according to an independent tracker, 51 OpenAI models were scheduled to be retired as of 17 October, and on 1 December older GPT Image models disappear from the API.
Also take note if you work via Azure OpenAI. It has its own dates: o1, o3-mini and o4-mini stop on 19 November, gpt-4o 2024-05-13 on 9 December.
Why small businesses often notice too late
In SMEs, AI rarely runs through one central system. It hides in integrations that were once built by a freelancer or agency: a chatbot on the website, a Zapier or n8n flow, a plugin, a background task. Such an integration often contains a fixed model ID. Nobody looks at it as long as it works.
On top of that, several schedules run side by side. OpenAI and Azure have different dates, and Anthropic and Bedrock or Google Cloud do too. So you need to know through which provider you actually purchase your AI. And sometimes there is no announcement at all: ChatGPT agent was removed from ChatGPT in early August 2026 without prior notice. Without an overview, your customers will be the first to notice.
Your checklist before 30 November
- Make an overview. List all AI integrations, with the provider and the model ID that is called for each one.
- Export what is disappearing. Are you using Agent Builder, Evals or saved prompts at OpenAI? Get your flows, evals and prompts out before 30 November.
- Test the replacement in time. A new model gives different answers and may also cost a different amount. Run your most important tasks side by side first and have a human assess the outcome.
- Choose a fallback model for each workflow. If one model is removed, you immediately know which one to switch to.
- Appoint one person responsible. One person per process who reads the emails about phase-outs and keeps the data up to date.
In addition, keep 12 November in mind: that's when Anthropic is adjusting its usage policy, which may affect how businesses deploy Claude. And structurally include supplier risk in your AI policy, so that this doesn't become an annual panic exercise.
At Silvar, we build AI solutions so that they don't depend on a single fixed model, with a human checking wherever that's needed. Unsure what runs in your integrations, or want a switch tested calmly? We're happy to help you look into it, well before the next deadline.