Harvey Launches Tenet, Open-Weight Legal AI Model
Summary
Harvey announced Tenet, its first post-trained open-weight legal AI model, built on the Kimi K3 base model in collaboration with Fireworks research. The model was trained using synthetic data, publicly available legal datasets, and expert human input, delivering nearly double the performance on LAB benchmarks for long-horizon legal tasks and a two-point gain in all-pass rate on LAB Contracts, while also achieving state-of-the-art results and cost efficiencies through reward shaping that favors concise, high-quality outputs. Tenet is aimed at legal workflows such as regulatory analysis, case law research, document review, and trial preparation, shows strong generalization on unrelated benchmarks like Mercor's APEX Agents and Crosby's Redline Bench, and supports Harvey's goal of enabling law firms to create proprietary AI models, with plans to scale the model, expand the Legal Agent Benchmark, and integrate Tenet into its broader AI suite, including a recent collaboration with Microsoft.
(Sourceļ¼Blockchain News)