Legal AI startup Harvey has introduced Tenet, its first proprietary model fine-tuned for legal work using the open-weight Kimi K3 from Chinese company Moonshot. This strategic shift moves Harvey away from dependence on external AI providers, including its investor OpenAI, replacing variable costs with owned technology optimized for the legal sector.
- Tenet is post-trained on Chinese open-weight model Kimi K3 from Moonshot
- Launch shifts Harvey from renting AI to owning and customizing its core technology
- Use of Chinese base raises supply chain and regulatory considerations for law firms
What happened
Harvey has officially launched Tenet, its first proprietary AI model developed specifically for legal work. The new model is post-trained on Kimi K3, an open-weight model developed by Chinese startup Moonshot, marking a strategic break from the previous reliance on external AI services from OpenAI, Anthropic, and Google. This move signals Harvey's intention to internalize a crucial piece of its identity and reduce dependence on vendors, including those invested in by itself such as OpenAI.
Tenet’s training involved extensive collaboration with legal professionals both in-house and from partner firms to create simulated disputes and case files that helped benchmark and improve the model’s legal reasoning capabilities. The launch represents a maturation of Harvey’s product offering in a competitive landscape where regional players, like European firm Legora, are also investing heavily in legal AI solutions.
Why it matters
Owning the AI engine behind its legal software transforms Harvey’s cost structure by moving from paying per-use fees to leveraging a fixed-cost asset. Each interaction routed through third-party models previously generated invoices from potential competitors, so building Tenet in-house enhances control and profitability. Additionally, the model is crafted specifically for legal tasks, promising increased quality and precision compared to generic AI services.
However, basing Tenet on a Chinese open-weight model presents nuanced challenges around licensing and compliance, especially for European law firms subject to the EU AI Act. Since Tenet is a post-trained version of Kimi K3 but may not reach thresholds for full regulatory responsibility transfer, users must navigate a complex chain of accountability starting from Moonshot’s technology in China. This scenario highlights emerging geopolitical and legal considerations in AI-driven professional services.
What to watch next
Industry observers and customers should monitor how Harvey manages licensing agreements with Moonshot, especially given revenue-based thresholds requiring separate contracts for large-scale model-as-a-service operators. Harvey’s reported run rate of over $350 million annualized places it well above such thresholds, making future negotiations and compliance critical to sustaining its AI deployment.
Additionally, the evolving regulatory environment, including the EU AI Act, will significantly impact law firms adopting Tenet or similar AI tools. How Harvey addresses transparency, responsibility, and data governance in light of its international technology supply chain could set a precedent for broader AI adoption in legal practice globally.