<p>Multi-autonomous underwater vehicle (AUV) cooperative hunting system hunts escaping targets through collaborative operations, which is used in military and civilian fields widely. However, in actual scenarios, the performance heterogeneity of hunter AUVs and escaping targets cannot be ignored, which pose great challenges to the cooperative hunting of multi-AUV. Therefore, this paper proposes a multi-AUV hunting algorithm based on performance level classification (MAHA_PLC) to improve the hunting efficiency of multi-AUV. First of all, this algorithm establishes a performance level classification strategy based on improved self-organizing map to determine the performance level and task type of hunter AUVs. This strategy establishes different mechanisms to evaluate the hunting behavior of hunter AUVs, change the task type and working area of hunter AUVs, and optimize neuron weight vectors, which can improve the allocation effect of hunting tasks. Then, this algorithm uses the random searching strategy to search escaping targets quickly. Furthermore, this algorithm designs a hunter alliance generation strategy based on trade-off performance elements to generate the hunting alliance of escaping targets, which establishes a hunting efficiency equation based on the movement speed and attack capability of escaping targets. Finally, this algorithm uses ant colony algorithm to design the hunting path for hunter AUVs. Compared to other hunting algorithms, the proposed MAHA_PLC can determine the task type of hunter AUVs and generate hunter alliance. Simulation results show that the average residual energy of hunting system is increased by 18%, and the average time for hunting escaping targets is reduced by 80%.</p>