• Picks
  • Discover
  • Alternatives
  • About

© 2026 Willet

  • Newsletter
  • About
  • FAQ
  • Terms
  • Privacy & Cookies
Join the Discord (opens in a new tab)
Harvey avatar

Harvey LAB

Benchmark tasks and harness for testing AI agents on legal work

  • Agent frameworks and tools
  • AI models and runtimes
View on GitHub
Website

Featured by Willet

What it is

Harvey LAB is a benchmark of about 2,000 legal tasks spanning 24 practice areas plus contracting, each with instructions, source documents, and grading rubrics. It ships an execution harness and grader, also installable as a Python package, that run agents in a sandbox, score their output with all-pass rubrics and an LLM judge, and produce reports and comparison dashboards. Researchers and legal AI teams use it to measure how well agents handle realistic assignments such as M&A data-room review.

Images

Facts

Stars
1.5k1,489

Forks
273
Commits
64
Contributors
21
Language
Python
License
MIT
Last commit
Oct 19, 2026
On GitHub since
March 2026
First release
September 2026
Platforms
CLI, library
Self-hostable
No
Tags
  • agent-harness
  • ai-agents
  • llm-evaluation

Last synced from GitHub on Oct 20, 2026

Share and more

Comments

No comments yet. Start the conversation.

The 10 best overlooked open-source projects, every Monday.

What people are saying

Updated Oct 20, 2026
  • Bluesky“that's what drew my eye there, the other scores here are single-digit percentages whilst Gemini busts onto the scene with a one-out-of-five pass rate—and…”Sep 30, 2026 · 4 likes · 1 reply (opens Bluesky in a new tab)
  • Bluesky“9. Sakana Fugu Orchestrates Multi-Agent Tasks Using Gemma 4 Conductor LINK 10. Harvey Labs Legal Agent Benchmark LINK”Aug 10, 2026 · 0 likes · 0 replies (opens Bluesky in a new tab)