# harveyai/harvey-labs

A benchmark built to evaluate and improve agent capabilities for supporting legal work.

Repository: https://github.com/harveyai/harvey-labs
Canonical: https://ross.abutalabs.com/products/harvey-labs
Homepage: https://www.harvey.ai/blog/introducing-harveys-legal-agent-benchmark
Language: Python
License: MIT
License Family: permissive
Last push: 2026-08-26T04:28:25+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 35, longevity 11
- inputs: {"age_days": 156, "days_push": 7, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1261, forks 215 (observed 2026-08-28T04:04:10.088230+00:00)

## What it is
Harvey LAB (Legal Agent Benchmark) is an open-source benchmark from Harvey AI for evaluating LLM agents on realistic legal work, spanning 24+ practice areas with over 1,600 tasks. It pairs a dataset of agent instructions, documents, and scoring rubrics with a Python execution harness that runs agents, scores them via all-pass rubric and LLM-judge evaluation, and produces comparison reports.

## Use cases
- benchmark LLM agents on legal tasks
- evaluate how well AI agents handle legal work like M&A due diligence
- find a dataset of legal tasks with rubrics for agent evaluation
- compare different models on legal reasoning and contract analysis
- run rubric-based scoring of agent outputs with an LLM judge
- measure agent capabilities for law firm workflows

## When to choose
- You need a standardized, citable benchmark to measure how LLM agents perform on realistic legal assignments across many practice areas
- You want an execution harness with all-pass rubric scoring and LLM judge behavior for evaluating agent runs
- You are researching legal-domain agent capabilities and want versioned, reproducible evaluation tasks and reports

## When to avoid
- You need a production legal AI assistant or document analysis product rather than an evaluation benchmark
- You want an agent framework for building or deploying agents rather than scoring them
- Your evaluation target is not legal work, in which case a general-purpose agent benchmark would fit better

## Facets
- artifact type: dataset
- maturity: active
- function: benchmarking, machine-learning
- domain: artificial-intelligence, large-language-models, legal
- platform: python, cli
- tags: llm-evaluation, agent-benchmark, legal-tech, rubric-scoring, llm-as-judge, evaluation-harness, task-dataset, ai-agents

## Member repositories
- harveyai/harvey-labs (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:10.088230+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T05:04:19.069786+00:00, confidence not recorded.
  - readme: https://github.com/harveyai/harvey-labs (fetched 2026-08-28T04:04:10.088230+00:00, sha b24296a88638)
  - homepage: https://www.harvey.ai/blog/introducing-harveys-legal-agent-benchmark (fetched 2026-08-29T12:16:55.043484+00:00, sha f1e787854fe6)
  - site_page: https://www.harvey.ai/en-US/about/law-schools (fetched 2026-08-29T12:16:55.058411+00:00, sha fa5f1bfac59b)
  - site_page: https://www.harvey.ai/en-US/company (fetched 2026-08-29T12:16:55.053570+00:00, sha 776e73fb3f6b)
  - site_page: https://www.harvey.ai/en-US/newsroom (fetched 2026-08-29T12:16:55.056020+00:00, sha eb0c00a8744b)
- Data as of 2026-08-30T08:39:29.467469+00:00.