You are comparing apple to oranges. Claude is better, sure, and I'd probably use it over deepseek but deepseek is an open model. For me, this makes deepseek quite superior (not from a benchmark/output perspective) to all the other closed models.