If the reasoning didn't need to be exposed to a user, are there any ways in which you get better performance or effect by using the same LLM methods, but using

If the reasoning didn't need to be exposed to a user, are there any ways in which you get better performance or effect by using the same LLM methods, but using a language better suited to that? (Existing language or bespoke.) (Inspired by movies and TV shows, when characters switch from English to a different language, such as French or Mandarin, to better express something. Maybe there's a compound word in German for that.)