(Image credit: Future)
If you've used ChatGPT, Gemini or Claude for any length of time, you've probably had the experiene of reaching for a model you've grown used to only to find it gone from the menu. This typically happens when a new model launches, but it could happen just days after a release. And while it feels like a deletion, it almost never is. What actually happens to an AI model when it's "retired" is actually pretty interesting. Old models get demoted or recycled into something that I'm calling "refurbished AI" because... there really isn't a name for it yet.Here's where the models goFirst, let's get to the bottom of what "retired" actually means. Every major lab runs a near-identical lifecycle, even if the labels differ slightly. Anthropic's is the clearest to follow. A model is Active while it's fully supported, becomes Legacy when it stops receiving updates, moves to Deprecated when it still works but is no longer recommended and has been handed a retirement date, and finally goes Retired, at which point requests to it simply fail.OpenAI uses similar language, distinguishing a "legacy" model that no longer gets updates from a "deprecated" one that has an official shutdown date on the calendar. Google, for its part, treats "deprecation" as the announcement and "shutdown" as the moment the endpoint is switched off for good.The key thing for everyday users to know is that most of these stages happen out of sight. By the time a model vanishes from your app, it has usually been winding down through this pipeline for weeks or even months.The afterlives of old models
