what language is most efficient in terms of tokens

A user observed that asking Deepseek to respond only in Chinese reduced daily token usage by 10-20%, despite the user not understanding the language. This led to a query about whether similar token savings could be achieved with Claude or other languages, suggesting an interest in optimizing output token usage across different AI models and languages.