Claude model retirement dates

The three states, as Anthropic defines them

An active model is fully supported. A deprecated model still works but is no longer recommended, and carries a replacement and, eventually, a retirement date. A retired model no longer answers: requests to it fail. Anthropic also uses "legacy" for a model that will receive no further updates and may be deprecated later — this site marks a couple of older models that way on their pages, while they remain active in Anthropic's table.

A floor, not a deadline

Every active model carries a "not sooner than" date. It is the earliest day that model can be retired, not the day it will be. Anthropic commits to notice before retiring a publicly released model — the lifecycle note under the first table states the minimum — so the practical reading of a floor is: after this date, a retirement announcement can arrive at any time, and your migration then has to fit inside the notice period alone. If your migration, testing included, takes longer than that, it has to start before the floor. The first table sorts models by floor, soonest first, and marks the ones already passed.

Retired here is not retired everywhere

These dates govern the platforms Anthropic operates — the Claude API, Claude Platform on AWS and Microsoft Foundry. Amazon Bedrock and Google Cloud keep their own schedules, and Anthropic's pricing page still lists several models there that are retired on the Claude API; the last column of the retired table shows which. A retired id that still answers through a partner cloud is running on someone else's calendar, not a reprieve you can plan around.

What to do with this page

Search your code and configuration for every model id you call, including ids that only appear in fallback lists. Check each against the tables, and for anything deprecated, retired or near its floor, follow its reference page to the named successor. The retired model id error page covers what a missed retirement looks like in production.