Abstract
The resolution of extensive-form zero-sum games is a fundamental challenge in computational game theory, addressed through various algorithms, each with unique strengths and limitations. This paper presents a comprehensive comparison of leading algorithms, using Poker-like games as benchmarks to assess their performance. For each algorithm, optimal parameters were identified, and evaluations were conducted based on exploitability, average utility, iterations per second, convergence speed, and scalability. The evaluation process comprised three stages. First, algorithms were tested on two-player variants of Kuhn, Leduc, and Royal Poker. Second, the scalability of the Kuhn Poker algorithm was examined by extending it to games with three to five players. Finally, convergence speed and scalability across all algorithms were systematically compared. The findings reveal significant trade-offs and performance distinctions, providing practical guidance for selecting algorithms suited to specific applications. This work advances the field by enhancing algorithmic understanding, refining evaluation methodologies, and offering valuable insights into the relative efficiency of strategies in multi-agent competitive environments.