Harvey Releases Tenet: A Kimi K3-Based Model Post-Trained for Long-Horizon Legal Agent Tasks

Harvey has released Harvey Tenet, a specialized legal AI model built by post-training the Kimi K3 base model using Fireworks AI infrastructure, targeting long-horizon agentic workflows in legal work. The model is designed for complex, multi-step legal tasks such as contract analysis, due diligence, and case research that require sustained reasoning across large document contexts. This release is notable as a concrete example of domain-specific post-training on a capable open base model to create a vertically specialized agent — a pattern increasingly viable for enterprise AI teams. Developers building legal tech or vertical AI agents can study this architecture: Kimi K3 as a strong reasoning base, combined with domain-specific RLHF or supervised fine-tuning via a scalable serving layer like Fireworks. It also signals that legal AI is moving beyond simple retrieval-augmented generation toward genuine long-horizon autonomous task execution.