Harvey's Legal Agent Benchmark
Can the model produce useful work in a professional knowledge-work setting?
Top rankings
One row per model, using its best reported score across effort settings.
| Rank | Model | Best score | Best reported setting |
|---|---|---|---|
| 1 | Grok 4.7SpaceXAI | 19.6% | xhigh effortSpaceXAI report1 value · 1 reportSep 21, 2026 · source |
| 2 | Gemini 3.8 FlashGoogle | 10.0% | Reported configurationGoogle DeepMind report1 value · 1 reportSep 2, 2026 · source |
| 3 | Gemini 3.7 FlashGoogle | 8.8% | Reported configurationGoogle DeepMind report1 value · 1 reportSep 2, 2026 · source |
| 4 | Claude Opus 5Anthropic | 6.7% | Reported configurationGoogle DeepMind report1 value · 1 reportSep 2, 2026 · source |
| 5 | Claude Sonnet 5Anthropic | 5.0% | Reported configurationGoogle DeepMind report1 value · 1 reportSep 2, 2026 · source |
| 6 | GPT-5.6 SolOpenAI | 2.5% | Reported configurationGoogle DeepMind report1 value · 1 reportSep 2, 2026 · source |
| 7 | GPT-5.6 TerraOpenAI | 0.8% | Reported configurationGoogle DeepMind report1 value · 1 reportSep 2, 2026 · source |
Effort curve
Every sourced cost-linked effort value for this exact version. Lines connect complete sweeps only.
Definition and comparison boundarypublic methodology
Can the model produce useful work in a professional knowledge-work setting?
Professional analysis, document, finance, office, or domain-specific tasks. Higher is better. The value is the percentage reported in this lab's table.
Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.
SpaceXAI reports 19.6% for Grok 4.7 at xhigh; the table labels this metric as legal work but does not print the harness or task revision. Complex legal workflows; the model card labels this metric All pass rate.
Method / source