Harvey announces Harvey Tenet, its first in-house, proprietary model for legal work, trained on mock disputes and case files using its own version of Kimi K3
Context & Ripple Effects
Harvey began as a legal-AI application company built on external foundation-model technology: its 2023 Series A backed a generative-AI tool for law firms, and its later Series B coverage identified OpenAI technology as part of the product stack. The company has since been described as an $11 billion legal-software business, making control of its model layer a material change in where it differentiates.
Tenet turns that application-layer history into a proprietary-model strategy, using mock disputes and case files alongside Harvey's version of Kimi K3. It follows the company’s progression from the early Sequoia-led financing for a law-firm AI tool toward building legal-specific intellectual property.
First-order effects
- Harvey gains direct control over a model trained for its legal workflows, rather than relying exclusively on third-party models for the core reasoning layer.
- Kimi becomes an upstream component of Harvey’s legal offering, while Harvey—not Kimi—owns the product relationship and legal-specific training approach.
Second-order effects
- Harvey can differentiate its legal product through its proprietary training data and evaluation work, shifting competitive emphasis from access to a general model toward performance on legal tasks.
- Kimi’s published API pricing provides a visible input-cost benchmark, but Tenet gives Harvey more scope to determine how that underlying model is packaged into its own service.
Third-order effects
- If legal-AI vendors follow Harvey’s path, the sector’s durable advantage will increasingly sit in proprietary domain training, workflow integration, and customer distribution rather than in reselling general-purpose model access.
- The move points to specialized software companies becoming AI-native systems integrators: assembling external foundation models while retaining ownership of the domain-specific layer customers use.
The trend: Vertical AI companies are moving from foundation-model dependence toward proprietary, domain-trained systems that control more of the product stack.