14 October 20262 min read

OpenAI is retiring GPT-4o mini: check your n8n workflows now

OpenAI will remove GPT-4o mini from the API on 14 January 2027. The Automation Vault uses that model, so its workflows need a controlled migration.

What to do now

On 14 October 2026, OpenAI deprecated GPT-4o mini and announced its API shutdown for 14 January 2027. The model is selected in the Automation Vault workflows, so do not wait for failed runs. Prepare the change now:

  • Search your n8n workflows for OpenAI Chat Model nodes using gpt-4o-mini or gpt-4o-mini-2024-07-18.
  • Duplicate each affected workflow before editing it. Keep the active version unchanged while you test.
  • In the duplicate, select OpenAI's recommended replacement, gpt-6-luna, if your n8n OpenAI node already offers it.
  • Run the duplicate with representative input and compare the structured output, classifications and drafted messages with the current workflow.
  • Check every expression that reads model output. Only activate the replacement after the downstream nodes still receive the expected fields.

OpenAI lists the dated GPT-4o mini snapshot among the models that will stop working on 14 January 2027. Its model page also marks GPT-4o mini as deprecated. A workflow that still calls it after the shutdown will fail at the model node rather than quietly choosing another model.

Why a simple model swap is not enough

The model name sits in one node, but its output controls the rest of the workflow. A replacement can phrase a category differently, omit a field when the prompt leaves room for interpretation, or produce longer text. That matters in workflows that route requests, write rows into a sheet or draft customer mail.

Test the cases that can cause damage, not only the happy path. For inbox triage, include an ambiguous message. For receipt capture, use a low-confidence document. For a routing agent, include a request that must be escalated rather than approved automatically. Compare both the visible answer and the data passed to the next node.

If the replacement is not available in n8n yet

Do not invent a model identifier and do not edit the live workflow. First update n8n to a current stable release, then check the model selector again. If your installed node still does not support the recommended model, leave the working workflow in place and schedule the migration before the January deadline. The deprecation is urgent because it affects every Vault workflow that calls the model, but the API does not shut down today.

This notice concerns the OpenAI API, not the earlier retirement of models inside ChatGPT. It also does not require changing prompts merely because a newer model exists. Change the model in a copy, test the outputs that your automation depends on, and then switch deliberately.

← All posts Every claim here links to its source.
Pre-launch

Hourbench builds the workflows these posts talk about.

The Automation Vault is 20 ready-to-run n8n workflows plus a bonus one. It is built and installable, and in final testing against real accounts before it goes on sale.

See what's coming →