Massive Legal Embedding Benchmark (MLEB)
The most comprehensive benchmark for legal embeddings.
Last updated — Highest score — 0 models
Datasets, paper, and citation
LRB
Legal RAG Bench
Isaacus Caselaw
4,976 rows
BEQ
Bar Exam QA
Stanford University Caselaw
350 rows
SC
SCALR
Faiz Surani and Varun Iyer Caselaw
763 rows
ER
ECHR Retrieval
Isaacus Caselaw
600 rows
SJK
Singaporean Judicial Keywords
Isaacus Caselaw
1,500 rows
GHR
GDPR Holdings Retrieval
Isaacus Caselaw
1,500 rows
ATGR
Australian Tax Guidance Retrieval
Isaacus Regulation
329 rows
ILS
Irish Legislative Summaries
Isaacus Regulation
1,500 rows
ULLT
UK Legislative Long Titles
Isaacus Regulation
234 rows
CCR
Contractual Clause Retrieval
Isaacus Contracts
225 rows
LTR
License TL;DR Retrieval
Isaacus Contracts
195 rows
CCQ
Consumer Contracts QA
Noam Kolt Contracts
478 rows
MLEB is described in our research paper on arXiv, including methodology and aggregate findings.
If you use MLEB in academic work, please cite the paper. Example (BibTeX):
@misc{mleb2025,
title={The Massive Legal Embedding Benchmark},
author={Isaacus},
year={2025},
eprint={2510.19365},
archivePrefix={arXiv},
primaryClass={cs.CL}
}Open evaluation code and raw results: github.com/isaacus-dev/mleb. Introductory post: Introducing MLEB.
Ready to build with the world’s best legal embedding model?
Read our docs, explore our models, and join our platform.