You should only use reasoning models if your prompt needs reasoning (eg debugging a weird error or writing an optimized algorithm).