DEV Community

Cover image for o3 retires on Azure 22 days before OpenAI's API. Same model, two shutdown dates.
Versyka
Versyka

Posted on AI-assisted

o3 retires on Azure 22 days before OpenAI's API. Same model, two shutdown dates.

Azure kills your o3 22 days before OpenAI does. Not December. Same model, two dates:

  • On OpenAI's API, o3-2025-04-16 shuts down on December 11, 2026.
  • On Azure OpenAI, the o3 (2025-04-16) deployment retires on November 19, 2026.

Also retiring on Azure on November 19:

Azure model (version) Suggested replacement
o1 (2024-12-17), o1-pro (2025-03-19) gpt-5.6-sol
o3 (2025-04-16), o3-pro (2025-06-10), o3-deep-research (2025-06-26) gpt-5.6-sol
o3-mini (2025-01-31), o4-mini (2025-04-16) gpt-5.6-terra

Why this catches teams

Most people track the model maker's deprecation page. But the cloud that hosts the model keeps its own calendar, with its own dates and its own replacements. Azure does, Amazon Bedrock does, Google Vertex AI does. Sometimes the cloud is later. This time it's three weeks earlier.

A second trap: on Azure, your code calls a deployment name you chose, not the model ID. Grepping for o3 finds nothing if your deployment is called reasoning-prod.

How to check

List every deployment with its model and version:

az cognitiveservices account deployment list \
  -g <resource-group> -n <account-name> \
  --query "[].{deployment:name, model:properties.model.name, version:properties.model.version}" \
  -o table
Enter fullscreen mode Exit fullscreen mode

Then compare each model and version with Azure's model retirement page, not OpenAI's.

Keep the host in your inventory

A model inventory that says "o3" isn't enough. It needs "o3, version 2025-04-16, on Azure, deployment reasoning-prod". The host decides the date.

Versyka's retirement calendar keeps cloud dates separate from the maker's (Azure, Bedrock and Vertex AI each get their own), so a model on Azure is checked against Azure's date.

Where do your models run: the maker's API, or a cloud?

Top comments (0)