Harvey, the U.S. legal AI unicorn, announced that it has developed Harvey Tenet, a legal-specific model, based on Moonshot AI's open-weight model Kimi K3 through post-training techniques.

This marks the first time a Chinese open-weight model has entered the training system of a leading U.S. vertical AI company.

Harvey previously relied primarily on closed-source models from OpenAI, Anthropic, and others to provide legal services. The company serves more than 2,400 institutions and over 200,000 lawyers, with clients including numerous world-class corporate legal departments. Public information shows its current valuation at $11 billion with annualized revenue of approximately $350 million, making it the highest-valued legal AI unicorn. It is now in the process of raising a new funding round at around $15.5 billion.

Kimi K3 is Moonshot AI's open-weight model. Open-weight means that once developers obtain the base model, they can use their own data for post-training to adapt the model to specific scenarios. Harvey Tenet is a legal-specific model post-trained on top of Kimi K3.

Legal AI demands extremely high levels of professionalism and compliance. Closed-source models can only be accessed via API, leaving developers unable to deeply modify model behavior. Open-weight models offer another path: obtain the base model, use vertical-domain data for post-training, and make the model better suited to business needs. For a company like Harvey serving over 200,000 lawyers, model controllability and customizability matter more than benchmark scores.

From relying on closed-source models to adopting a Chinese open-source base, Tenet's rollout reflects that open-weight models are gaining substantial trust in vertical tracks. The fact that a U.S. AI company valued at tens of billions of dollars is willing to put a Chinese open-weight model into its core training system already speaks volumes.