Legal Research Bench: Results as Published

Legal Research Bench tests whether an AI agent can answer realistic US legal research questions, using tools such as case-law search, web search and document retrieval to produce a supported answer. The tasks span eight practice areas: Administrative/Regulatory, Business & Commercial, Civil Litigation, Constitutional/Civil Rights, Criminal, Family, Health and Immigration. The questions, gold-standard answers and rubrics are written and peer-reviewed by practicing lawyers, and each question has 1 to 31 rubric items (mean 9.35). An LLM judge grades each response against the rubric; the primary metric is all-pass accuracy, where a question counts only if every rubric item is met, and the page also reports a partial-credit score by share of rubric points. Vals AI publishes the leaderboard.

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.