Legal Research Bench: Measuring End-to-End Reliability in Long-Horizon Legal Research Agents

Legal research is a core legal workflow, but reliability remains the central barrier to automating it. Lawyers must identify controlling authority, verify that it remains valid, reconcile statutes and cases, and synthesize a grounded answer. Language-model agents are a natural fit for this retrieval-intensive workflow, but a single missing authority, stale citation, or wrong legal conclusion can make an otherwise plausible answer unusable. We introduce Legal Research Bench (LRB), an expert-authored benchmark of 413 open-ended United States legal-research questions, each paired with a gold answer, verified supporting authorities, and a binary grading rubric. We evaluate thirteen frontier models in a harness with web search, case-law search, page parsing, and retrieval/state tools. We score agent responses through all-pass grading with source verification, where a response is correct only if every required criterion is satisfied and its cited authorities verify. We also validate the LLM judge against expert attorneys to ensure that benchmark scores align with attorney judgment. Current agents remain far from reliable: the strongest model, Claude Opus 4.8, satisfies every required rubric criterion on 42.9% of questions. Performance also varies substantially by task setting: all-pass rates differ across areas of law, are lower on questions requiring reconciliation of conflicting authorities, and are not reliably improved by more turns, tool calls, or inference cost.