Legal Agent Benchmark all-pass rate — Harvey held-out set (LAB all-pass (Harvey held-out))
We show this table for reference; we do not rank on it.
Harvey AI's strict held-out task success rate requiring every rubric criterion to pass.
Benchmark score on LAB all-pass (Harvey held-out) — September 27, 2026
We compile the LAB all-pass (Harvey held-out) rows from provider self-reports. Claude Opus 5 leads the table at 11.7%, followed by Claude Opus 5.5 (8.3%). We do not use these results to rank models overall.
Claude Opus 5
Anthropic
Claude Opus 5.5
Anthropic
2 modelsAgenticCurrentDisplay onlyUpdated September 27, 2026
Benchmark score table (2 models)
ScoreAbout LAB all-pass (Harvey held-out)
Year
2026
Tasks
Harvey-held-out legal-agent tasks
Format
All-criteria pass rate
Difficulty
Professional legal work
Section 8.13.3 separately reports Harvey's evaluation on its held-out set. It is not merged with Anthropic's internal-harness result.
Freshness and provenance
Version
LAB all-pass (Harvey held-out) 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does LAB all-pass (Harvey held-out) measure?
Harvey AI's strict held-out task success rate requiring every rubric criterion to pass.
Which model scores highest on LAB all-pass (Harvey held-out)?
Claude Opus 5 by Anthropic currently leads with a score of 11.7% on LAB all-pass (Harvey held-out).
How many models are evaluated on LAB all-pass (Harvey held-out)?
2 AI models have been evaluated on LAB all-pass (Harvey held-out) on BenchLM.
Compare top models on LAB all-pass (Harvey held-out)
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 2,000+ readers.
One email each week. Unsubscribe anytime.