Harvey unveils its own legal-focused LLM, Tenet, to cut costs and boost control

Harvey announced Tenet, its inaugural in-house large language model designed for legal applications, marking a shift from using external AI services. The model aims to automate tasks that typically require hours of lawyer time, offering a lower-cost alternative to platforms such as OpenAI and Anthropic, which are increasingly targeting the legal market. To build Tenet, Harvey enlisted attorneys to generate and evaluate mock disputes, using the data to train a version of Moonshot's open-source Kimi K3 model.

Alongside Tenet, the firm is rolling out a "Memory" capability that lets users store workflow preferences for future interactions. Harvey says it will soon release research comparing Tenet's performance to other models, though the system is not yet deployed and rollout dates remain undisclosed. Founder Gabe Pereyra envisions Tenet eventually serving as a foundation for law firms to fine-tune their own AI assistants, potentially reshaping Harvey's role from software vendor to professional-services partner.