Skip to content

Legal AI

Harvey LAB

Long-horizon legal work with documents and per-task rubrics; Harbor snapshot has 1,251 tasks.

About Harvey LAB

What it measures

Rubric-criterion pass rate and all-criteria task pass rate.

Scope & limitations

Publisher-created benchmark: disclose provenance. Main repository and third-party ports differ in task counts and expansions; freeze the chosen release.

Evaluation environment

Legal agent with document/file tools and fixed rubric judge.

Source

Original benchmark

Dataset & task catalog ↗