This isn’t technically true. Most model providers don’t send the thinking tokens anymore, so if you switch from one provider to another, you will be missing large parts of the conversation.