Every AI model retires. Here’s when.

A provider’s notice that a model is on its way out. After deprecation the model usually can’t be used for new work, and on its retirement date calls to it stop working. Providers set these dates on their own schedule, often within a year or two of a model’s release.

Legacy means the model still works but the provider has scheduled its end. Deprecated means no new access, while existing use keeps working until retirement. Retired means calls to the model fail. Not every provider uses all three stages, so some models go straight from available to retired.

Open “Add to calendar” and pick Google Calendar, Apple Calendar, Outlook, Outlook.com or Microsoft 365, or copy the feed link into any calendar app that supports subscriptions. To follow one provider, filter by it first and the menu switches to that provider’s feed. Subscribed calendars update on their own when a date changes.

From official data published by AWS Bedrock, Azure AI Foundry, Google Vertex AI and OpenAI. Draftt’s analysts validate every date before it appears here, so the schedule reflects what each provider has actually committed to.

Draftt’s analysts review provider changes regularly, and the page and calendar feeds pick up each validated change on their own.

Requests to it fail, so any feature, agent or pipeline still calling it breaks. Move to the provider’s recommended replacement before the retirement date, and test for differences in output quality, limits and cost along the way.

That’s the hard part: models get called from many services, keys and accounts. Draftt AI Governance finds every model your org calls across providers, puts each one on its provider’s clock, and shows the owners and workloads that have to move before the date.