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

Research introduces 'Legal Research Bench' (LRB), a new benchmark designed to measure the end-to-end reliability of language model agents in complex legal research workflows. The benchmark comprises 413 expert-written U.S. legal research questions, each with a gold answer, supporting authorities, and a binary grading rubric. This initiative aims to quantify the reliability of frontier models for legal tasks, where accuracy is paramount due to the high cost of errors.

The reliability of automated legal research tools is a critical factor for their adoption and utility in the legal sector. This benchmark provides a standardized method for evaluating AI performance, enabling organizations to assess the true value and risks associated with integrating language models into high-stakes legal workflows. Understanding model reliability directly impacts decisions regarding resource allocation, technological investment, and operational efficiency.

Legal research is a core, time-consuming workflow for legal professionals.

Forward consideration, not a verified fact.