Harvey Launches Tenet, a Legal Model Post-Trained on Moonshot AI's Kimi K3
The AMW Read
Tenet is an incremental update to Harvey's already covered in-house model effort, but its open-weight post-training strategy materially informs legal AI differentiation.
Named counterparties: OpenAI, Moonshot AI
Harvey Launches Tenet, a Legal Model Post-Trained on Moonshot AI's Kimi K3
Harvey has introduced Tenet, a proprietary legal model built through post-training on Moonshot AI's open-weight Kimi K3. The company said Tenet uses roughly 1,750 legal-task environments that convert lawyers' factual judgments, citation requirements, risk ratings, and delivery formats into scored training signals. The source reports that the initial post-training run used about 150 Nvidia B300 GPUs over two months; Tenet remains in research preview. This extends AI Market Watch's August 20 coverage of Tenet's inclusion in the Harvey II platform.
The strategic point is less that a legal software company released another model than that it is trying to own the layer between a general model and a lawyer's finished work product. Harvey's reported token consumption rose from about 1 trillion monthly tokens in January to 14.5 trillion in June, making model access a material operating cost as agentic legal work expands. A specialized evaluation environment and expert feedback loop could become a stronger defense than prompts or a generic retrieval layer, while reducing dependence on a single closed-model supplier.
For builders, the transferable asset is not merely choosing an open model: it is creating repeatable, expert-defined task environments that can measure whether outputs meet professional standards. Investors should treat reported benchmark results as early evidence, then test whether Tenet improves customer outcomes, gross margins, and switching costs when deployed across real law-firm workflows.
