One and the same text in Russian takes about twice as many tokens as its English equivalent. This means that the context window fills up faster, tokens are consumed more quickly, and if you’re using the API, it all costs more.
A simple solution is to talk to your Claude Code or Codex in English. Even if your English isn’t perfect, the LLM will understand you just fine. And the savings on tokens will be quite noticeable.