Harvey, the US legal AI startup worth $11 billion, just unveiled its first proprietary legal model. The foundation? China's open-source Kimi K3, from Moonshot AI.
Harvey has raised more than $1 billion in total funding and, until now, relied on general-purpose models from OpenAI and Anthropic. Tenet changes that: it's a model Harvey can call its own, post-trained specifically for legal reasoning.
To get there, Harvey had lawyers write up hypothetical disputes and case materials, then score the model's legal reasoning. That data went back into training Kimi K3. Harvey says Tenet already performs on par with top general-purpose models across major legal benchmarks — at a lower cost. Just don't ask for the receipts yet: no benchmark scores have been published, and Tenet isn't live.
Next up, Harvey wants law firms to fine-tune their own specialized legal models on top of Tenet.