Were you using v2?

Pydantic docs do clearly state that multple levels of nesting of Pydantic objects can make it much slower, so it isn't particularly surprising that such models were slow.