爱思助手苹果版安装

how does deepseek r1's performance in math-heavy benchmarks compare to gpt-4o