BigLaw Bench Workflows: SPA Deal Points

Harvey is expanding its publicly-released BigLaw Bench data to include workflow datasets for evaluating legal workflow agents, starting with extracting deal points from Share Purchase Agreements (SPAs).

• Harvey's SPA agents achieve 98.47% accuracy on this task, compared to 66.04% (GPT-4o) and 72.27% (Gemini) for general-purpose foundation models.
• The company plans to release more workflow datasets and incorporate these workflows into its Assistant and Vault products.