Harvey nearly doubles legal benchmark performance with expert post-training

Legal AI firm Harvey post-trained an open-source model with expert data from Mercor, nearly doubling task completion on its held-out legal benchmarks. Mercor argues expert feedback is uniquely valuable because models that produce flawed legal memos also fail to spot the errors.

2026-11-11 ~ 2026-11-11 · 2 related posts

• Why expert feedback matters: an AI that writes a weak legal memo can't spot its flaws — ccerrato147 · 2026-11-11
• Harvey post-trained an open model with expert data, nearly doubling completed legal tasks — ccerrato147 · 2026-11-11