The security and stability of the Internet heavily rely on the reliability of inter-domain routing, with the Border Gateway Protocol (BGP) playing an important role. However, due to its lack of built-in security mechanisms, BGP remains vulnerable to a wide range of routing anomalies such as hijacking and leak, which can result in severe service disruptions and substantial economic losses. Although numerous BGP anomaly detection approaches have been proposed, the absence of standardized datasets, unified evaluation criteria, and sustainable assessment mechanisms has significantly limited the practical impact of these methods. In this paper, we propose RADBench, a comprehensive and reproducible benchmark designed to systematically evaluate the performance of BGP anomaly detection methods. We construct the largest public dataset of real-world BGP anomalies, introduce multidimensional evaluation criteria, and build a modular benchmarking framework that supports sustainable evaluation. Experimental results demonstrate the benchmark’s ability to reveal critical differences in detection performance across methods and offer actionable insights for real-world deployment.