Harvey and Applied Compute Post-Train Legal Model, Pushing Quality-Cost Frontier

ypatil125 · x · 2026-08-15

Harvey and Applied Compute post-trained a model for Harvey's Review Table, significantly improving the quality-cost Pareto frontier. Key methods include building a corpus of public legal documents, using oracle agents to generate expert-reviewed query-answer-citation tuples, designing a reward function with sentence-level citation precision/recall, and leveraging domain expertise from Harvey's legal researchers.

Related event: Harvey Post-trains GLM-5.2 to Cut Legal AI Costs by 50%(4 posts)→

Original post →

More from Companies & People

Companies & People channel →