Legal Agent Benchmark (Harvey Held-Out): leaderboard
Harvey legal-agent held-out evaluation using closed-universe matter files and expert rubrics, scored by all-pass task success.
Metric: All-Pass Rate (%). Source: www-cdn.anthropic.com. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 13.3 | #16 |
| 2 | Claude Opus 4.8 | 10.4 | #37 |
| 3 | GPT-5.5 | 2.1 | #31 |
| 4 | Gemini 3.5 Flash | 0.8 | #60 |
| 5 | Gemini 3.1 Pro | 0 |
Interactive version: theaggregate.ai/benchmark?slug=legal-agent-benchmark-harvey-held-out · How It Works · Data refreshed daily, snapshot 2026-10-20.