9 records · Browse by task

Legal benchmarks

Find tests for legal reasoning, research, case retrieval, and document work.

Jurisdiction, task set, tools, and grading change what each result means. These tests do not establish legal accuracy in a particular matter or replace a review of your workflow.

Records keep metrics and harness variants separate. Source rows can include different model and agent configurations; linked profiles count distinct public model identities. Neither count represents independent evaluations. Dates describe the stored source record.

Public model rows count finite, displayable results linked to reviewed original artifacts. Publisher comparison charts can include other providers' models. Follow the result link to inspect each row's source and run conditions.

Legal reasoning and research

PRBench Legal
41 stored source rows · 22 linked model profiles
Snapshot label: October 15, 2026 snapshot
Original source
Published professional-reasoning configurations remain separate from legal-agent runs.
41 published rows
LegalBench
151 stored source rows · 85 linked model profiles
Source update in snapshot: 2026-10-13
Source snapshot captured: 2026-10-15T18:47:28.403Z
Original source
The Vals-hosted LegalBench result stays separate from any other LegalBench harness.
Legal reasoning task views
Legal Research Bench
78 stored source rows · 65 linked model profiles
Source update in snapshot: 2026-10-12
Source snapshot captured: 2026-10-15T18:47:28.403Z
Original source
US-law research task views preserve strict and weighted-pass results.
US-law legal research tasks

Legal agent work

Harvey's Legal Agent Benchmark
79 stored source rows · 66 linked model profiles
Source update in snapshot: 2026-10-12
Source snapshot captured: 2026-10-15T18:47:28.403Z
Original source
The Vals-hosted Legal Agent Benchmark preserves task-level and criterion-pass views.
Legal agent work across documents, spreadsheets, presentations, and files
LAB all-pass (Anthropic harness)
1 public model rows linked to original artifacts
Original source
All-pass grading under Anthropic's internal harness.
1,235 legal-agent tasks
LAB criterion-pass (Anthropic harness)
1 public model rows linked to original artifacts
Original source
Criterion-pass grading under Anthropic's internal harness.
1,235 legal-agent tasks
LAB all-pass (Harvey held-out)
3 public model rows linked to original artifacts
Original source
All-pass grading on the Harvey held-out task set.
Harvey-held-out legal-agent tasks
LAB criterion-pass (Harvey held-out)
3 public model rows linked to original artifacts
Original source
Criterion-pass grading on the Harvey held-out task set.
Harvey-held-out legal-agent tasks

Historical evidence

CaseLaw v2
54 stored source rows · 35 linked model profiles
Archived source board
Source update in snapshot: 2026-05-04
Source snapshot captured: 2026-10-15T18:47:28.403Z
Original source
Archived Canadian case-law questions remain useful context for that specific retrieval task.
Canadian case-law question answering

Other collections

All benchmarks