This paper introduces a new benchmark for testing community detection algorithms, which accounts for the heterogeneity in the distributions of node degrees and community sizes, unlike traditional benchmarks. The authors tested two popular methods: modularity optimization and Potts model clustering. The new benchmark is more challenging for algorithms, revealing limitations that may not be apparent in simpler tests. The benchmark graphs are constructed with power-law distributions for both node degrees and community sizes, allowing for a wide range of network structures. The results show that modularity optimization has a resolution limit, making it difficult to detect small communities. The performance of both methods depends on parameters such as the average degree and the mixing parameter, which affects the algorithm's ability to identify communities. The new benchmark is suitable for testing algorithms because it can generate large networks quickly and covers a wide range of sizes. The authors conclude that the new benchmark is essential for evaluating the accuracy and efficiency of community detection algorithms, as it reflects the complexity of real networks.This paper introduces a new benchmark for testing community detection algorithms, which accounts for the heterogeneity in the distributions of node degrees and community sizes, unlike traditional benchmarks. The authors tested two popular methods: modularity optimization and Potts model clustering. The new benchmark is more challenging for algorithms, revealing limitations that may not be apparent in simpler tests. The benchmark graphs are constructed with power-law distributions for both node degrees and community sizes, allowing for a wide range of network structures. The results show that modularity optimization has a resolution limit, making it difficult to detect small communities. The performance of both methods depends on parameters such as the average degree and the mixing parameter, which affects the algorithm's ability to identify communities. The new benchmark is suitable for testing algorithms because it can generate large networks quickly and covers a wide range of sizes. The authors conclude that the new benchmark is essential for evaluating the accuracy and efficiency of community detection algorithms, as it reflects the complexity of real networks.