Harvey Legal Agent Benchmark

Open-source long-horizon legal agent benchmark where agents work from client-matter instructions, closed-universe matter files, and expert rubrics to produce reviewable legal work product.

6rows
all_pass_task_successprimary metric
2026-05-28sampled

Metadata

ID
harvey_lab
Category
Legal
Release
2026-05-06

Metrics

All-Pass Task Success

Showing 2 latest source slices.

Latest Results

Provider-published system-card benchmark scores parsed from Anthropic's Claude Opus 4.8 capability evaluation tables. Rows are marked self-reported and should be interpreted as source claims unless independently reproduced.

Rank Subject All-Pass Task Success Model Match Provenance Sampled
1 Claude Opus 4.8 9.6% Claude Opus 4.8
anthropic-claude-opus-4.8
 		Self-reported 2026-05-28
1 Claude Opus 4.7 7.1% Claude Opus 4.7
anthropic-claude-opus-4.7
 		Imported 2026-05-28
2 Claude Sonnet 4.6 5.4% Claude Sonnet 4.6
anthropic-claude-sonnet-4.6
 		Imported 2026-05-28
3 Claude Opus 4.6 4.2% Claude Opus 4.6
anthropic-claude-opus-4.6
 		Imported 2026-05-28
4 GPT-5.5 2.1% GPT-5.5
openai-gpt-5.5
 		Imported 2026-05-28
5 Gemini 3.5 Flash 0.8% Gemini 3.5 Flash
google-gemini-3.5-flash
 		Imported 2026-05-28