Harvey Launches Tenet, Open-Weight Legal AI Model

Harvey has introduced its first post-trained open-weight model, Harvey Tenet, which the company says achieves significant performance improvements in long-horizon legal tasks while maintaining cost efficiency. The announcement, made on August 20, 2026, marks a new milestone in Harvey’s ongoing effort to advance legal artificial intelligence tailored for law firms and legal teams.

Tenet builds on the Kimi K3 base model and was developed in collaboration with Fireworks research. Harvey’s post-training process incorporated synthetic data, publicly available legal datasets, and human expert inputs, yielding what the company describes as a nearly twofold improvement in LAB benchmarks for long-horizon legal tasks. On LAB Contracts, Tenet raised the all-pass rate by 2 percentage points compared to the base K3 model. Notably, it also achieved state-of-the-art performance on LAB Contracts and placed second across broader LAB benchmarks.

Beyond performance, Harvey emphasized Tenet’s cost-efficiency. Open-weight models inherently offer lower token costs, but Harvey’s post-training approach further optimized for efficient token usage during inference. By employing reward shaping to favor concise, high-quality outputs, the company claims it managed to improve both quality and cost stability.

The model is tailored for demanding legal use cases, including regulatory analysis, case law research, document review, and trial preparation. Harvey Tenet also demonstrates strong generalization across unrelated legal benchmarks such as Mercor’s APEX Agents and Crosby’s Redline Bench, despite not being directly trained on these tasks. This suggests robust transfer capabilities, a critical feature for AI systems aiming to handle diverse legal workflows.

Harvey’s broader research agenda focuses on enabling law firms to build their own specialized AI models. This includes tools to help firms retain and utilize their proprietary knowledge securely. The company is also expanding the Legal Agent Benchmark (LAB), its in-house standard for evaluating legal AI performance, to cover more jurisdictions and practice areas. These efforts align with Harvey’s strategy to scale its compute infrastructure and bring advanced research models into production-grade applications.

While Tenet is not a publicly traded asset, its potential implications for legal AI are significant. By lowering operational costs and improving task performance, models like Tenet could reshape how law firms approach high-volume, complex legal work. Harvey’s next steps include scaling Tenet’s capabilities and integrating them into its broader suite of legal AI solutions, as announced in its recent collaboration with Microsoft.

For law firms exploring AI adoption, Tenet represents a step toward more accessible and efficient legal intelligence tools. The question now is whether Harvey can maintain its trajectory as it scales this technology to meet the demands of real-world legal practice.

Image source: Shutterstock