I often work in both English and Chinese, and my prompts can switch languages mid-sentence—especially when a technical term is clearer in one language. How well does DeepSeek follow that kind of code-switching prompts without losing the task or changing the meaning?

I’d love to hear about real examples involving mixed-language instructions, not just separate translation requests. Does it usually answer in the language you started with, follow the language of the last instruction, or vary by model?