Legal Research Bench
Legal Research Bench evaluates tool-using agents on questions involving statutes, regulations and case law across US federal and state jurisdictions. The published harness includes web search, page parsing, stored-document retrieval and CourtListener search.
Reported results
| Muse Spark 1.3 MaxHarvey | 55.29 | |
|---|---|---|
| Claude Opus 5Anthropic | 55.29 | |
| Claude Fable 5.1Anthropic | 55.29 | |
| Claude Fable 5Anthropic | 49.52 | |
| GLM-5.3Zhipu AI | 49.04 | |
| Grok 4.6xAI | 48.08 | |
| GPT-5.6 SolOpenAI | 48.08 | |
| Qwen 3.8 MaxAlibaba | 47.6 |
The top three tie at 55.29%. Claude Opus 5 reaches 90.58% under weighted partial credit but only 55.29% when every required item must pass. Conflicting-authority questions reduce scores by 6–17 points per model.
What the benchmark measures
Legal research questions across eight practice areas that require agents to find and combine sources. Answers are assessed for substance and supporting authority.
The evaluated unit is an agent or completed task. Read the source for the exact prompt, tool and harness conditions.
How it is scored
Two measures: questions passing every required item, and weighted scores that award partial credit.
Scores remain in the original unit. They are not normalised or combined with results from other benchmarks.
Sources
Under the strict measure, every required item must pass. The public leaderboard is updated separately from the original open-source release.