Harvey Tenet: Legal AI Model Cuts Costs & Boosts Profit

Legal AI startup Harvey has unveiled its first in-house large language model called "Harvey Tenet." This move aims to control costs and business direction, as Harvey previously relied on external models from companies like OpenAI and Anthropic. Here's the thing: Harvey's expenses grew quickly as its usage increased, because it had to pay providers every time lawyers used those external models. By developing Tenet, Harvey plans to shift more work to an internal engine, reducing external model fees and improving profitability without raising customer charges. What's interesting is that the launch coincides with large general-purpose AI companies entering the legal market directly. This trend raised questions for Harvey about competition from its own model suppliers. Gabe Pereyra, a Harvey co-founder, stated that quality, alongside cost, was a key reason for developing Tenet. Harvey also created its own legal reasoning data to train the model, hiring lawyers to design hypothetical disputes and case files. Tenet was trained on a low-cost open-source model called "Kimi K3" from Chinese startup Moonshot, then tailored with Harvey's legal data. The bottom line: Tenet is part of a larger product revamp called "Harvey 2," which also includes a new memory feature. This could allow lawyers to handle work that used to take hours or days, at a lower cost.