CanLegalRAGBench: Evaluating Retrieval-Augmented Generation on Canadian Case Law

In three linesCanLegalRAGBench is an evaluation benchmark for RAG systems applied to Canadian law, based on realistic queries and expert-annotated answers. The study shows open-source embedding models are competitive with closed-source alternatives, but identifies hallucinations in 8-29% of generated answers unsupported by retrieved documents.

Summary generated by Claude — human-verified