Unveiling the Mathematical Reasoning in DeepSeek Models: A Comparative Study of Large Language Models