# Legal Research Bench ![Category](https://img.shields.io/badge/Category-Legal%20AI-1f6feb) ![Metric](https://img.shields.io/badge/Metric-%25%20Accuracy-2da44e) ![Saturation](https://img.shields.io/badge/Saturation-Low-d29922) ![Created](https://img.shields.io/badge/Created-2026-555) > **Quick answer:** Legal Research Bench tests AI models on legal research tasks — case-law lookup, statutory interpretation, and citation accuracy — evaluating whether a model can produce reliable, well-cited legal research output. Kimi K3 scores 44.2% as of July 2026. ## At a Glance **What it tests:** A model's ability to conduct accurate legal research, including correctly citing case law and statutes and interpreting legal text. **Why it matters:** Legal research demands precise citation and interpretation; errors (including fabricated case citations) carry serious professional risk. This benchmark targets that specific reliability requirement. **Known limitations:** As an emerging benchmark, exact task composition and jurisdiction coverage are not yet independently published outside its citation by Moonshot AI. ## What Legal Research Bench Measures Legal Research Bench evaluates a model's ability to conduct legal research tasks — finding and citing relevant case law and statutes, interpreting legal language, and producing accurate legal analysis. It complements broader legal knowledge-work benchmarks (like Harvey Lab-AA) by focusing specifically on research and citation accuracy. ## Benchmark Specifications | Field | Value | |---|---| | Task category | Reasoning / legal domain | | Metric | % accuracy | | Saturation | Low | | Created by | Not yet independently documented | ## How Legal Research Bench Is Scored Models complete legal research tasks, scored on % accuracy against expert-verified reference answers, including correctness of legal citations and interpretations. ## State-of-the-Art Results | Rank | Model | Score | Source | Date | |---|---|---|---|---| | 1 | [Kimi K3](/models/moonshot/kimi-k3) | 44.2% | [Kimi K3 technical report](https://www.kimi.com/blog/kimi-k3) | 2026-07 | *Score sourced from Moonshot AI's Kimi K3 announcement, July 2026.* ## Legal Research Bench on Benchgen > No Benchgen results yet — [be the first to run Legal Research Bench](/benchmarks/platform/legal-research-bench). ## Legal Research Bench vs Other Benchmarks | Benchmark | What it tests | Saturation | |---|---|---| | **Legal Research Bench** | Legal research & citation accuracy | Low | | [Harvey Lab-AA](/benchmarks/platform/harvey-lab-aa) | Legal knowledge-work quality | Low | | [DeepSearchQA](/benchmarks/platform/deepsearchqa) | Multi-hop web research & synthesis | Low | | [ResearchRubrics](/benchmarks/platform/researchrubrics) | Rubric-graded research report quality | Low | ## Run Legal Research Bench on Your Model Benchgen lets you run Legal Research Bench against your own model, tracking legal research accuracy over time before deploying to legal-domain use cases. ## Frequently Asked Questions

What is Legal Research Bench?

Legal Research Bench is a benchmark testing AI models on legal research tasks, including case-law lookup, statutory interpretation, and citation accuracy.

What does a good score look like on Legal Research Bench?

Kimi K3 reports 44.2% as of July 2026, reflecting the significant remaining difficulty of fully reliable legal research for current frontier models.

Who created Legal Research Bench?

Legal Research Bench's originating team is not yet independently documented outside of its citation in Kimi K3's July 2026 technical report.

Legal Research Bench — Results

Legal Research Bench — Evaluations