According to marktechpost, legal AI company Harvey unveiled its first post-training model, Harvey Tenet, on August 20, 2026, currently available as a research preview. Built on the Kimi K3 foundation model, Tenet was trained using asynchronous reinforcement learning techniques for long-duration legal tasks. The training data included synthetic data, publicly available legal data, and human expert data, with no customer data used. In Harvey's Legal Agent benchmark tests, Tenet demonstrated exceptional performance, completing nearly twice as many tasks as the base K3 model and achieving industry-leading results in the LAB: Contracts evaluation.