DistDebug-Bench: Can LLM Agents Diagnose Bugs in Distributed Systems?
DistDebug-Bench, a benchmark for evaluating agents'ability to diagnose complex-system failures at the code level, curates 120 high-quality real-world bugs from 11 widely used distributed systems, spanning 5 programming languages and 10 failure patterns.