Data as of Sep 26, 2026 · Based on 4,029,442 AI responses across 13,338 prompts · See how Parse measures this
LegalBench-RAGgithub.comLegalBench-RAG is an information retrieval benchmark designed to evaluate retrieval systems on complex legal contract understanding, enabling deterministic precision and recall down to exact character ranges. It provides a corpus and benchmark format (data/corpus and data/benchmarks) along with tooling to download, reproduce, or regenerate benchmarks that map queries to ground-truth snippets within the corpus. Benchmark generation relies on LLMs and requires agreeing to usage policies of sources like ContractNLI, CUAD, MAUD, and PrivacyQA, with scripts to generate and run the benchmark.
Parse Score
No contexts measured yet.
Excerpts where LegalBench-RAG appeared in the AI's answer
LegalBench-RAG focuses specifically on retrieving precise relevant passages rather than merely retrieving whole documents
LegalBench-RAG: This is a specialized benchmark and framework designed specifically to evaluate how well retrieval systems pinpoint exact legal references.
