ExplainBench introduces an automatic benchmark for evaluating explanations produced by coding agents. It tests whether explanations accurately describe both the intended behavior of buggy code and the effects of applying an agent-generated patch. The benchmark is based on a question-answering intuition: informative explanations should allow an LLM to answer relevant questions correctly. The authors report that explanation quality is a distinct evaluation axis, with ExplainBench ranking agents differently from SWE-bench Verified. The supplied abstract is truncated, so details of the question suite, models, scores, and deeper breakdown are unavailable here.
No heat snapshots are available in the last 24 hours.