Traditional UAV swarm task planning methods struggle to understand complex task instructions in dynamic environments, with prominent issues such as insufficient adaptability and low human-computer interaction efficiency. This paper proposes a UAV swarm task planning method based on a large language model (LLM), which deeply integrates the natural language understanding ability of LLM with existing multi-agent task allocation and route planning algorithms. Firstly, LLM is used to automatically extract key task planning elements from instructions and input them into the HATP planner to automatically complete multi-UAV task decomposition and allocation. Then, LLM is utilized again to extract route planning elements from the task allocation results, and automatically set the parameters of the route planning algorithm to generate the trajectory of each UAV. In addition, the time logic in the task planning results is analyzed, and a unified UAV scheduling table is established. The segment speed allocation of each UAV is determined according to the actual route of each UAV, so as to realize the time coordination of the overall task. Finally, the effectiveness of the proposed method is verified by typical multi-UAV task simulations.