Harvey Launches Harvey II and Tenet — Its First Proprietary Legal AI Model

Harvey released Harvey II on August 18 alongside Tenet, its first proprietary model post-trained end-to-end for legal reasoning on top of Moonshot AI's open-weight Kimi K3. Harvey II restructures the platform around matter-centric Spaces — persistent environments where documents, permissions, tasks, and history travel together as work moves between agents and lawyers — and a Memory layer that learns individual drafting style, citation preferences, and matter context, carrying those forward across Harvey, Word, and Outlook. The strategic significance of Tenet is straightforward: owning the model removes per-token inference costs from Harvey's P&L for core legal workflows and, critically, enables firm-specific model customization at scale. Building on Chinese open-weight infrastructure will draw scrutiny from enterprise security teams at confidentiality-sensitive buyers, but Harvey's calculated bet is that frontier-level benchmark performance at open-source inference economics changes the unit economics conversation with procurement.