The model ID you're calling has been retired

The error

A request that has worked for months suddenly starts failing with no code change on your end at all — the model id it targets has been retired, and the API simply stops accepting it past that date. This is one of the few errors on this list that isn't caused by anything in your request; it's caused by time passing.

Why this catches teams by surprise

Model retirement dates are published ahead of time, but a retirement date announced when a model first ships is easy to lose track of by the time it actually arrives, especially for a model that was working fine and never prompted anyone to go looking for news about it. Code that pins a specific model id directly, rather than routing through something that gets actively maintained, has no mechanism to notice a retirement is coming — it just keeps working until the date passes, and then it doesn't.

What "retired" actually means here

A retired model id isn't deprecated-but-still-working — deprecation is the warning period; retirement is the id no longer resolving at all. Some retirement dates pass without an explicit confirmation that the cutover executed exactly on schedule, so if you're relying on a model id you haven't touched in a while, don't assume "it worked last time I checked" still holds — verify against the current model lineup rather than your own last observation of it.

The fix

Migrate to the current model in that same tier — every retired model id in the current lineup has a documented successor, and the migration is rarely a drop-in replacement given how much can shift between generations on pricing, cache minimums and effort support. Check the retired model's own reference page for its named successor, and treat the move as a real migration rather than a one-line id swap: see what actually changed from Opus 4.6 to Opus 5 for the kind of thing that quietly changes underneath an id swap like this.