I’m trying to understand whether giving DeepSeek more room to reason changes not just answer quality, but also how consistent the answer is when I run the same prompt several times. My test cases are short logic and planning questions where small changes in the result matter.

Have you compared repeated runs at different reasoning budgets or settings? I’m interested in whether more reasoning reduces variation, simply produces longer explanations, or makes little difference once the prompt is clear.