The paper introduces a new class of benchmark graphs designed to test community detection algorithms more effectively than traditional benchmarks. The authors, Andrea Lancichinetti, Santo Fortunato, and Filippo Radicchi, address the limitations of existing benchmarks, such as the Girvan-Newman (GN) benchmark, which do not accurately reflect the heterogeneity of real networks in terms of node degrees and community sizes. The new benchmark graphs are constructed to mimic real-world networks by incorporating power-law distributions for both node degrees and community sizes. The authors test two popular community detection methods—modularity optimization and Potts model clustering—using this new benchmark. The results show that the new benchmark poses a more severe test to algorithms, revealing limitations that are not apparent with standard benchmarks. Specifically, the benchmark highlights the resolution limit of modularity optimization and the impact of graph size and link density on algorithm performance. The paper emphasizes the importance of realistic benchmarks in evaluating community detection algorithms and provides a software package to generate the benchmark graphs.The paper introduces a new class of benchmark graphs designed to test community detection algorithms more effectively than traditional benchmarks. The authors, Andrea Lancichinetti, Santo Fortunato, and Filippo Radicchi, address the limitations of existing benchmarks, such as the Girvan-Newman (GN) benchmark, which do not accurately reflect the heterogeneity of real networks in terms of node degrees and community sizes. The new benchmark graphs are constructed to mimic real-world networks by incorporating power-law distributions for both node degrees and community sizes. The authors test two popular community detection methods—modularity optimization and Potts model clustering—using this new benchmark. The results show that the new benchmark poses a more severe test to algorithms, revealing limitations that are not apparent with standard benchmarks. Specifically, the benchmark highlights the resolution limit of modularity optimization and the impact of graph size and link density on algorithm performance. The paper emphasizes the importance of realistic benchmarks in evaluating community detection algorithms and provides a software package to generate the benchmark graphs.