In this paper, a novel Gyro Fireworks Algorithm (GFA) is proposed by simulating the behaviors of gyro fireworks during the display process, which adopts a framework of multi-stage and multiple search strategies. At the beginning of the iteration, the gyro fireworks are full of gunpowder; they move via Lévy flight and spiral rotation, and the sprayed sparks are widely distributed and more balanced, which is an effective global exploration method. In the later iteration stages, due to the consumption of gunpowder, the gyro fireworks gradually undergo aggregation and contraction of spiral rotation, which is conducive to the search group to exploit the local area near the global optimal position. The GFA divides the iterative process into four phases, and each phase adopts a different search strategy, in order to enhance the diversity of the search of the population and to balance the exploration capability of the gyro fireworks search group in the global space and the exploitation of the local space. In order to verify the performance of the GFA, it is compared with the latest algorithms, such as the dandelion optimizer, Harris Hawks Optimization (HHO) algorithm, gray wolf optimizer, slime mold algorithm, whale optimization algorithm, artificial rabbits optimization, in 33 test functions. The experimental results show that the GFA obtains the optimal solution for all algorithms on 76% of the functions, while the second-placed HHO algorithm obtains the optimal solution for all algorithms on only 21% of the functions. Meanwhile, the GFA has an average ranking of 1.8 on the CEC2014 benchmark set and 1.4 on the CEC2019 benchmark set. It verifies that the GFA proposed in this paper has better convergence performance and better robustness than the competing algorithms. Moreover, experiments on challenging engineering optimization problems confirm the superior performance of the GFA over alternative algorithms.