A lawyer at a major firm drafts a contract clause, and a piece of software rewrites it in seconds. Another sifts a mountain of documents for the one passage that matters. Both are increasingly the work of Harvey, the legal-AI startup that on September 9 said it had raised $550 million at a valuation of $15.5 billion.
The round was led by Diffusion and Lightspeed, and it nearly doubles the company’s worth from the $11 billion it was valued at in March. As recently as December 2025, Harvey was valued at $8 billion. The climb has been steep and fast, and it has pulled in more than $1.55 billion in total funding.
The money arrives as Harvey begins to change what it sells. Two weeks before the round, the company released its first self-developed model, called Tenet, built on the open-weight model Kimi K3 and post-trained on legal data. The inference that runs it is provided by Fireworks.
That detail matters. Most AI startups rent frontier models from OpenAI or Anthropic and wrap a product around them. Harvey is now encouraging its customers to post-train open models themselves, betting that law firms want control over the models they rely on rather than a dependency on a handful of AI labs.
The legal industry has been a natural early adopter. Law firms handle enormous volumes of text, and much of the work of junior associates, reading, summarizing, comparing, drafting, is exactly what language models do well. Harvey has ridden that fit to a client list that includes many of the largest firms in the country.
But the industry’s attitude toward AI is complicated. Firms use the tools heavily, yet they are wary of handing their most sensitive client material to a handful of closed-model labs they cannot inspect. Harvey’s answer has been to make that wariness into a business, offering models the firms can see, fine-tune and run with a degree of independence.
The strategy carries risk. Building and maintaining your own model is more expensive than reselling someone else’s, and the frontier labs are improving quickly. Harvey is betting that in law, where trust and control are everything, a model a firm can shape will beat a model it merely rents, even if the rented one is smarter out of the box.
The valuation reflects that bet as much as the revenue. Harvey’s backers are paying for the prospect that it becomes the operating system for legal work, a category that spends hundreds of billions of dollars a year on labor. A $15.5 billion price is small against that pool, the argument goes, if Harvey can keep the largest firms on its platform.
Analysts said the rapid valuation climb mirrors the broader AI market’s willingness to pay up for enterprise software that shows real usage. Harvey has pointed to revenue growth and a growing roster of large customers, and the round’s structure, with two lead investors rather than one, suggests strong demand for the shares.
Harvey was founded in 2022 by Winston Weinberg, a former lawyer, and Gabriel Pereyra, a former research engineer, who built the product around the tasks lawyers actually perform. That grounding in the workflow of a firm, rather than a generic chat window, is a large part of why the largest firms adopted it so quickly.
The company has also expanded beyond drafting and research. It has moved into areas such as litigation analysis, regulatory work and the tools corporate legal departments use every day. Each new product deepens the data it holds and the switching cost for a firm that has woven Harvey into its workflow.
Competition is not standing still. A wave of startups has targeted the legal market, and the large tech platforms are pushing their own assistants into the same firms. Harvey’s defense is specialization: a product built for law, trained on law, and now owned end to end, from the model to the interface.
What Harvey does with the new capital will define the next phase. The company is expected to spend on further model development, on engineering that makes its tools faster and more accurate, and on the enterprise sales that win big-firm contracts. The days of renting a model and adding a logo are over for it.
The open-model approach also positions Harvey for a market that may be reshaped by regulation. If large firms and their clients come to demand that AI tools be auditable and controllable, a company that lets customers own and tune their models has an answer the closed labs may struggle to match.
For now, the round is a signal. In nine months, Harvey’s value has nearly doubled, and a startup that began as a thin layer over someone else’s model now ships its own. Legal AI has gone from a curiosity to a category, and Harvey has become its most valuable name.
