We've stopped upgrading the models of our classification workflows for >1 year at this point, meaning they're running acceptably on early/mid-2025 models. That said I believe there is a long tail of non-production-ized users who throw this into their everyday LLM chats.