To solve the UAV offline and online path planning problem, an adaptive cuttlefish algorithm with joint modification is proposed considering that the traditional cuttlefish algorithm may be trapped in local optimum and has low precision after long search. The regulatory mechanism combined with chaos perturbation and mutation learning is proposed to strengthen local search to improve search precision. Adaptive weight mechanism is introduced to diminish search space and ensure search efficiency. The auto filter mechanism based on individual fitness was used to adjust population diversity and to break away from local optimum. The proposed algorithm was firstly verified by tests of six benchmark functions, and then verified by path planning simulations in different environments. For offline path planning, success rate of the proposed algorithm reaches up to 100%, and paths planned by the proposed algorithm were closest to global optimum. The average track length of the proposed algorithm could be 7.3 units shorter than traditional cuttlefish algorithm, and it could reach 28.3 units shorter than particle swarm optimization. The simulation results show that the global search performance and the search precision of the proposed algorithm were significantly improved. Furthermore, as the environment became more complex, the improvement effects were more remarkable. For online path planning, the global path planning was transformed to the planning of several segmental paths, and a heuristic method was used to select the segmental goal. Simulation results show that the proposed algorithm met the real-time requirement and has high precision. The effectiveness of the proposed algorithm was further verified.