17 OpenAI Models to be shut down this Friday. Here's how to check your codebase in 10 minutes.
OpenAI is slated to shut down 17 model IDs on Friday, 23rd October. This isn't just a deprecation notice - once a model is retired, API requests to its ID can fail and should not be assumed to work with any given replacement.
If you pinned a model to use in 2024 for reproducibility, or have an automation in n8n, Make or Zapier that makes a call to one of these models, you're in danger of it breaking.
It can be easy to miss out on a scheduled task or background queue that fails repeatedly until noticed.
The 11 base models listed on the 23rd of October
| Model ID | Proposed replacement |
|---|---|
| gpt-3.5-turbo-0125 and gpt-3.5-turbo | gpt-5.6-terra |
| gpt-4-0613 and gpt-4 | gpt-5.6-sol |
| gpt-4-1106-preview | gpt-5.6-sol |
| gpt-4-turbo-2024-04-09 and gpt-4-turbo | gpt-5.6-sol |
| gpt-4.1-nano-2025-04-14 and gpt-4.1-nano | gpt-5.6-luna |
| gpt-4o-2024-05-13 | gpt-5.6-sol |
| gpt-image-1 | gpt-image-2.5 |
| o1-2024-12-17 and o1 | gpt-5.6-sol |
| o1-pro-2025-03-19 and o1-pro | gpt-5.6-sol |
| o3-mini-2025-01-31 and o3-mini | gpt-5.6-sol |
| o4-mini-2025-04-16 and o4-mini | gpt-5.6-terra |
The original list of models also includes six fine-tuned variants. Make sure to check your fine-tuned model IDs individually rather than making assumptions about changing the base name being enough.
Two traps that could break your migration
- Don't check only dated model IDs.
Your code might be calling o4-mini rather than o4-mini-2025-04-16. Make sure to search for both forms, including any aliases that might exist in environment variables and within your workflow.
- Similar names don't necessarily share a retirement date.
The supplied schedule differentiates between gpt-4o and gpt-4o-mini, and between gpt-image-1 and gpt-image-1.5.
This is important - a careless find-and-replace can miss out on any calls affected or modify models which are currently in service.
Make sure to confirm the actual retirement date for each model before changing production code.
The 10-minute check
Search through these locations:
The application source code and shared libraries
.envfiles and deployment configurationn8n, Make and Zapier workflow exports
Cron jobs, scheduled tasks and background workers
API request logs and monitoring dashboards
Fine-tuned model configurations
Don't just rely on a single text search - make sure to match both complete model IDs and any aliases to avoid false positives. Then verify each affected call with the official retirement schedule, select an appropriate supported replacement and test the migrated request.
A replacement isn't always equivalent - make sure to check response formats, tool calling, image inputs, latency, cost and output quality before shipping the change.
### I've built a free scanner for this
If you'd rather not inspect everything manually, try out Dead Model Check.
Paste code, configuration or logs, and it'll identify model ID matches and associated retirement information.
It runs in your browser (your pasted content is not uploaded by this scanner). The goal is to make model migrations easier to plan, including any future retirement dates, rather than finding a broken API call after a deadline.
Need help with the affected code?
If the scan identifies retired model calls in your production code, and you don't have the time to fix them, I offer a fixed-fee migration service.
$149 - Model Migration and Verification
The scope is simple: identify the affected calls, update them to appropriate supported models and verify the changes. The exact scope of work would depend on your codebase and the number of affected integrations.
More information is available on the [Dead Model Check page].
*### One more thing to check
*
Model retirements aren't the only source of breakage. Deprecation of API endpoints can require significantly more work than just replacing a model ID.
If your application uses the Assistants API, make sure to check its official retirement schedule separately. Migration of a model name won't fix an API that's itself been retired.
The takeaway: spend ten minutes identifying the affected model calls now and then verify the dates, migrate deliberately and test before the deadline.
Always confirm retirement dates and replacement models against the official OpenAI documentation before deploying any changes.
Top comments (1)
when you run the 10‑minute scan, do you also flag model usage embedded in Dockerfiles or just runtime imports?