I’m testing a workflow where the instructions are in English, the source text is in another language, and the examples include code identifiers that shouldn’t be translated. DeepSeek sometimes seems to handle each part well on its own, but I’m unsure how dependable it is when all three appear together.

For anyone using DeepSeek mixed-language prompts, have you noticed particular trouble with code-switched text, technical vocabulary, or keeping names unchanged? I’d appreciate examples of what you tested and which model or interface you used.