Existing path planning methods of the unmanned aerial vehicle (UAV) rely heavily on complex obstacle-avoidance logic and are trained on the precise and static environment. This process makes it difficult to generalize to unknown scenarios with dynamic disturbances. To this end, we propose GPT4-o and Environment-aware Driven Planning Path for Unmanned Aerial Vehicle (termed as GED), which achieves the hierarchical path planning methods including global and local path plannings. Specifically, we first construct a virtual scene in Unity, the top-down view of the virtual scene then is taken as the input for the vision-language large model Molmo to further obtain the coordinates of buildings within the virtual scene. The prompts related to the building coordinates are fed into GPT-4o to generate multiple global path planning results that connect buildings based on the given home and goal, where the path with the shortest Euclidean distance is considered as the optimal global path. Moreover, the local path planning utilizes the proximal policy optimization combining with the real-time object detection to refine the final global path when UAV approaches building on the final global path, further generating the refined final path. Through extensive experiments, GED achieves excellent performance on randomly generated buildings in Unity. The code will be released soon.