Harvey AI Secures $550M to Expand Legal Automation Platform and Custom Model Development

Harvey AI Corp. has announced a $550 million funding round valuing the company at $15.5 billion, marking its second major capital raise within six months. The round was led by Diffusion and Lightspeed Venture Partners, with participation from more than a dozen additional investors including Sequoia, Kleiner Perkins and Goldman Sachs.

The startup operates a cloud-based platform designed to help legal professionals streamline repetitive tasks. Its customer base spans 80% of the top 100 U.S. law firms by ranking, and the company counts half of the Fortune 10 among its enterprise legal department clients.

Platform Capabilities

Harvey's system allows users to maintain a document repository called Vault that can hold up to 100,000 files. An integrated AI-powered search function identifies relevant patterns and connections within stored materials. Attorneys can query the system to locate specific contract types or clauses that require updates in response to regulatory changes, for instance.

When drafting agreements, legal professionals frequently need to examine prior contracts to identify language worth incorporating into new deals. Harvey's technology retrieves matching documents from Vault while also pulling in external information such as legal precedents and statutory language.

The company recently introduced AI agents capable of handling more sophisticated workflows. These agents can analyze large document sets for potential complications—such as in due diligence reviews for investment firms—while requesting clarification from attorneys when necessary.

Custom Language Model Launch

Coinciding with the funding announcement, Harvey unveiled Tenet, its proprietary large language model built on Kimi K3, an open-source LLM containing 2.8 trillion parameters. The model incorporates 896 specialized neural networks, each trained for particular task categories.

Harvey trained Tenet using legal documents and implemented a custom harness—a set of prompts and technical components meant to enhance output quality. According to the company, Tenet achieves 20% better performance than Kimi K3 on certain contract processing assignments and surpasses Fable 5 and GPT-5 Sol across multiple benchmarks.

The company indicated that it will invest fresh capital in computing resources to support ongoing AI research efforts. "New generalist models" represent a particular focus area for development going forward. Building proprietary models requires substantial upfront investment but can reduce long-term operational expenses by decreasing reliance on third-party model services, particularly since inference typically consumes a larger share of AI workload costs than model training.

Legal AI Benchmark

Harvey introduced LAB alongside Tenet, a benchmark designed to evaluate how well language models perform on legal tasks. The current version encompasses approximately 1,200 distinct tasks, with plans to extend coverage across additional jurisdictions and practice areas.