In order to address the limitations of conventional Particle Swarm Optimization (PSO) algorithms, including premature convergence and reduced performance in high-dimensional and complex UAV path planning environments, this paper proposes a semantically enhanced UAV path planning algorithm based on the DeepSeek (LLM-PSO). Firstly, the traditional velocity-based update mechanism of PSO is replaced by a position generation process using natural language prompts through the DeepSeek API, significantly enhancing the global exploration capabilities and optimization performance. Secondly, an output parsing and verification mechanism is integrated to ensure the physical feasibility and safety of generated solutions. Comparative experiments conducted in both simple and complex scenarios demonstrate that the proposed LLM-PSO algorithm exhibits superior optimization performance, faster convergence speed, and better robustness compared to traditional PSO and its improved variants.