Most startups are told to pick one narrow task and do it better than anyone. Winston Weinberg and Gabriel Pereyra were told the same thing and went the other way. Their bet, as Harvard Law School’s Center on the Legal Profession put it in an April 2026 essay, was that “the models are the product.”
Two roommates and a language model
Weinberg was a securities and antitrust litigator at O’Melveny & Myers. Pereyra had been a research scientist at DeepMind, Google Brain and Meta AI. According to TechCrunch, which profiled the pair when Harvey raised its Series C in 2024, Pereyra showed Weinberg OpenAI’s GPT-3 and Weinberg saw it could improve legal workflows. LawNext’s Bob Ambrogi, in a January 2026 interview, describes them as roommates who started the company from a shared apartment while Weinberg was still a first-year associate.
Harvey launched in 2022. Its first large proof point was a law firm, not a software buyer. Allen & Overy, now A&O Shearman, trialled the product from November 2022, and by February 2023 it was rolling Harvey out to more than 3,500 lawyers across 43 offices, according to coverage citing the Financial Times and LawNext. The firm’s trial logged about 40,000 queries on day-to-day client work. David Wakeling, who led the firm’s Markets Innovation Group, called it a “game-changer.”
The decision to go broad
Harvard’s essay describes the choice that shaped everything after. Rather than start with a single task such as contract review, Harvey built a general-purpose assistant for lawyers, then layered on what large firms demanded before they would trust it: security, isolated environments, admin controls and separation between client matters. Later came workspaces and workflow tools, and in December 2025 a shared space where firms and their clients work together.
Investors followed the traction. In July 2024, TechCrunch reported a $100 million Series C led by GV at a $1.5 billion valuation, with total funding at $206 million. Harvey told the outlet its annual recurring revenue and its headcount had both tripled since the previous December. Then the rounds accelerated. TechCrunch’s September 2026 report puts the valuation at about $8 billion that December and $11 billion in March, when Harvey raised $200 million.
September 2026: $550 million, and a different bet
On September 9, 2026, Harvey announced $550 million at a $15.5 billion valuation. Diffusion and Lightspeed Venture Partners co-led. Sapphire Ventures and Whale Rock joined as new investors, alongside existing backers including Sequoia, Kleiner Perkins, Andreessen Horowitz, Coatue and GIC. TechCrunch counts more than $1.55 billion raised in total and notes the valuation nearly doubled in about nine months.
Harvey’s announcement says 80% of the Am Law 100 use the product, along with in-house teams including five of the Fortune 10. It did not disclose revenue, and neither did TechCrunch’s report, so this profile gives none. Third-party estimates of Harvey’s annual recurring revenue conflict, and we have left them out.
The more telling detail is what came just before the round. A few weeks earlier Harvey released its first in-house model, Harvey Tenet. TechCrunch reports it was built on the open-weight model Kimi K3, post-trained on legal data with help from the inference provider Fireworks. Harvey also launched Harvey LAB, a benchmark for legal agents, and the announcement is headlined around helping legal teams “own their intelligence.” Harvey encourages customers to post-train open-weight models of their own.
For a company whose first product ran on GPT-3, that is a notable turn. The founders’ own line in the announcement is about the customer, not the model: “The opportunity for companies to accelerate their competitive advantage with AI has never been higher.”
What to take from it
- Breadth can be the wedge. Harvey skipped the single-task start-up playbook and won by being the general tool a firm could adopt once and extend.
- Trust is the product in regulated work. Security and client separation, not model quality alone, is what the Harvard essay says let Harvey into the largest firms.
- Own the layer your customers care about. Moving from rented models toward post-trained open-weight ones changes who controls cost and data.
Related founders
- Michael Truell and the $60 Billion Cursor Deal — another AI company that scaled on top of other people’s models.
- Anton Osika Built Lovable from Stockholm — a fast-compounding AI valuation from a different angle.
- Christina Cacioppo Took Vanta from a Spreadsheet to $300M ARR — selling trust-heavy software to cautious enterprise buyers.
Winston Weinberg’s and Gabriel Pereyra’s backgrounds and the 2022 launch are per TechCrunch’s July 23, 2024 report on Harvey’s Series C, which also supplies that round’s size, valuation, lead investor, total funding, and the tripling of ARR and headcount. The roommate account is per LawNext (Bob Ambrogi, January 20, 2026) and Harvard Law School’s Center on the Legal Profession (Anthea Roberts and David Wilkins, April 3, 2026), the latter also the source of the “the models are the product” quote and the description of Harvey’s product evolution. Allen & Overy’s trial and rollout figures, and David Wakeling’s quote, are per legal.io’s February 17, 2023 report citing the Financial Times and LawNext. The September 9, 2026 round, its investors, customer figures and the founders’ quote are per Harvey’s own announcement, opened directly; the prior valuations, total funding, Harvey Tenet and Kimi K3 details are per TechCrunch’s September 9, 2026 report. No revenue figure appears because Harvey did not disclose one in the announcement and third-party estimates conflict.
