Benchmarks / Legal work

Legal work

Legal research, drafting, and review.

Part of Legal.

• APEX-Agentsby Mercor, Box and HarveyProject level · Pass@1

  82.4%Gemini 4 Argon (high) · 10 Oct 2026

  Strong
• Harvey Legal Agent Benchmarkby Harvey and Vals AITask level · All-pass rate (held-out set)

  25.4%Muse Spark 1.2 · 22 Sep 2026

  Emerging
• Legal Research Benchby Vals AITask level · All-pass accuracy

  55.3%Muse Spark 1.3 Max · 22 Sep 2026

  Emerging
• RedlineBenchby Crosby Legal and micro1Task level · Turn-weighted rubric score

  50.5%GPT-5.5 · 16 Jun 2026

  Emerging