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
Harvey built an $11 billion legal-software business on top of other companies' AI models. Now it's trying to prove it can build one of its own.
Context & Ripple Effects
Harvey previously built its legal AI products on OpenAI technology, including when it raised an $80M Series B for its OpenAI-based legal platform. Subsequent fundraising reports tracked a business scaling its legal-software ambitions rather than building a foundation model from scratch.
Tenet marks a change in where Harvey owns differentiation: it is developing a legal-specific model trained on mock disputes and case files, while still building on its version of Kimi K3.
First-order effects
- Harvey gains direct control over the legal-model layer of its product, giving its law-firm customers a Harvey-developed system rather than a service wholly dependent on third-party models.
- Kimi becomes an explicit part of Harvey's model stack, while Harvey's training materials and legal specialization become the product-specific component.
Second-order effects
- Third-party model providers lose some control over Harvey's end-user differentiation as Harvey shifts value toward its own legal training and evaluation work.
- Legal-AI rivals that rely on general-purpose models face added pressure to show comparable domain-specific performance, not merely access to an underlying model.
Third-order effects
- The move points to a layered legal-AI market in which foundation models remain infrastructure while vertical software companies compete on proprietary data, training, and workflow fit.
- If this approach spreads, legal-software valuations will hinge less on model access and more on whether vendors can turn specialized legal materials into defensible model behavior.
The trend: Vertical AI companies are moving from reselling general-purpose models toward owning the domain-specific model layer that shapes customer outcomes.