Harvey Open-Sources 100M+ Token Synthetic Law Firm Dataset for Agent Evaluation

HazyResearch · x · 2026-08-08

Legal AI company Harvey, in collaboration with EngramLab, has open-sourced a synthetic law firm dataset containing over 100 million tokens.

The dataset simulates work products from 250+ legal matters across 46 clients, spanning approximately 10,000 files. It serves as an evaluation environment to test AI agents' ability to search and understand a firm's past practice to inform present work, mimicking the institutional knowledge of a tenured associate or partner.

Original post →

More from Research

Research channel →