“Most of our accounts grow pretty massively,” said Winston Weinberg, co-founder and CEO of Harvey. “You’ll sell to a Comcast or to a law firm, and they’ll buy a couple hundred seats, and then they expand that usage pretty quickly.”
That customer behavior explains how Harvey, an AI startup focused on legal work, reached $100 million in annual recurring revenue in just under three years. From a prototype coded in a San Francisco apartment to one of the most widely adopted generative AI platforms in the legal sector, Harvey’s ascent has been marked by rapid customer expansion, targeted enterprise use cases, and a strategic alignment with OpenAI that began before ChatGPT ever went viral.
Founded in late 2022, Harvey now counts more than 500 enterprise customers in 53 countries. Its clients include Paul Weiss, Allen and Overy, WongPartnership, A&L Goodbody, Comcast, PwC, and the Adecco Group. It has been adopted across law firms, corporate legal departments, and legal services organizations. In August 2025, Harvey crossed the $100 million ARR mark, up from $75 million just four months earlier. The company had ended 2023 with $10 million in ARR, according to GetLatka, marking a more than 500 percent year-over-year growth in 2024.
How They Got to $100 Million ARR
Harvey’s growth has been unusually fast, even by generative AI standards. The company’s revenue climbed from zero at launch in November 2022 to $10 million ARR by the end of 2023. The real acceleration came in 2024, when revenue reached $65.8 million, followed by a sharp uptick in the first half of 2025 that brought the total to $100 million by August.
The pace reflects a few deliberate choices.
First, Harvey priced like an enterprise platform from day one. Customers typically started with 100 to 500 seats, often expanding within months. This initial land-and-expand motion was not driven by pilots or free trials, but by paid deployments inside legal departments and top-tier law firms. Expansion happened quickly once internal teams validated use cases across practice areas.
Second, the company built a sales team that mirrored its buyers. Many of Harvey’s sales and success leaders are former attorneys from firms like Latham and Watkins, Skadden, and Wachtell. That team was able to meet general counsels and partners at the right level, with an understanding of workflows, risk profiles, and firm economics. When combined with a high-performing product, that credibility created short sales cycles and high retention.
Third, the product itself was never positioned as an experiment. From the beginning, Harvey was pitched as a production-grade tool capable of real legal output. This positioning helped the company avoid the platform fatigue that many generative AI tools encountered during the first wave of adoption.
Fourth, Harvey focused on organizations that had both high volume and high value legal work. That includes not just global law firms but in-house legal teams at large corporations. Legal departments at Fortune 500 companies became natural fits given their need for scalable research, compliance, and contract automation.
Finally, the company’s deep integration with OpenAI gave it early access to foundational models and a path to differentiate. By co-developing a legal-specific model with OpenAI, Harvey avoided the performance issues that plagued off-the-shelf solutions and positioned itself as a defensible platform.
This revenue trajectory mirrors other AI-first companies like Arcads and Instantly.ai that have scaled quickly on the back of focused use cases. But unlike consumer or prosumer tools with small average contract values, Harvey’s enterprise motion was built around larger, longer-term contracts with real revenue scale.
A Cold Email That Turned into a Strategic Partnership
The company’s founding story has become a reference point for how timing and proximity to foundational AI research can create leverage. Weinberg and Pereyra began working on a prototype after experimenting with OpenAI’s GPT-3. At the time, Weinberg was working 80-hour weeks as a securities litigator. Pereyra had spent years building large language models. They tested an early demo that answered legal questions pulled from Reddit and cold-emailed OpenAI’s general counsel.
That email led to a July 4 meeting with OpenAI’s executive team and eventually resulted in Harvey becoming the first startup backed by the OpenAI Startup Fund. Over the next five months, they built Harvey’s first legal agent out of their apartment and launched the company in November 2022 with five employees and no revenue.
Legal Trust, Not Just Legal Tech
Harvey succeeded where many legal tech startups failed—by solving the problem of professional trust. Its partnership with OpenAI extended beyond access to models. Pereyra worked directly with the team to develop a custom legal model trained on 10 billion tokens of legal data, starting with Delaware case law and expanding to all US jurisdictions.
When tested with lawyers from ten major firms, 97 percent preferred the output from the Harvey-trained model over base GPT-4. That performance was not just about accuracy. It was about familiarity with legal formatting, citation logic, and the nuanced reasoning style expected in legal memoranda and briefings.
Harvey’s adoption has mirrored the land-and-expand model often seen in enterprise SaaS. According to internal data, the company grew from 40 customers at the beginning of 2024 to 235 by the end of the year. That number surpassed 500 by August 2025. The weekly average number of users has quadrupled in the past 12 months.
A Second-Generation Legal AI Rollout
Charles Russell Speechlys, a UK-based law firm, recently selected Harvey as its enterprise-wide generative AI solution after replacing Springbok AI, a legal tech provider that was acquired earlier this year by Cleary Gottlieb. The firm evaluated multiple platforms across nine categories. Harvey ranked first in every one.
“It scored 8.4 out of 10 overall, and 8.7 out of 10 in one,” said Sam Cohen, who led the evaluation. “The next best was 7.4 and the others were in the sixes. These are people who lived on Springbok, so there was no question about quality or functionality. They unequivocally said Harvey.”
CRS is now rolling Harvey out across all legal professionals globally, integrating it with both iManage and LexisNexis, and migrating 26 previously built Springbok bots into Harvey.
Weinberg called the selection a “comprehensive evaluation” and said CRS’s feedback-first approach aligned closely with Harvey’s core product principles.
Expansion into India and Europe
Harvey is now building beyond its initial footprint. It will open a new office in Bengaluru, India, where engineering, sales, and operations teams will begin work later this year. Weinberg has publicly stated that the long-term goal is for every lawyer in India to use the platform. Bengaluru was chosen for its combination of legal sector depth, technical talent, and status as a major technology hub.
The company has also announced a new office in Germany as part of its broader European expansion strategy.
With more than 340 employees and recent executive hires including former Twitter engineering director Siva Gurumurthy as Chief Technology Officer and ex-Stripe leader John Haddock as Chief Business Officer, Harvey is investing heavily in global scale-up efforts.
Beyond Document Generation
Harvey’s most ambitious product work is centered around legal agents that can handle entire workflows autonomously. The company is experimenting with custom reasoning models for more complex legal tasks such as M&A deal analysis, discovery review, and billing operations. Some firms have begun co-developing software with Harvey and are reselling tools to their clients under revenue-sharing agreements.
There is also growing interest in automating legal back-office functions. Firms have approached Harvey for billing automation and spend management tools. While the core product remains focused on legal professionals, partnerships in operational areas may expand over time.
The economics of high-compute AI make sense in this domain. A motion that costs $250,000 to draft manually can be evaluated at scale with token-based inference, even if each task costs thousands of dollars to run. Weinberg has made it clear that Harvey is betting on long-term improvements in inference cost, but for now, the legal ROI already justifies the infrastructure spend.
Harvey is not attempting to replace lawyers. Its platform is built to elevate their capabilities and automate the repetitive, mechanical parts of legal practice. The company’s leadership has studied the failures of past legal tech ventures like Atrium and Clearspire, and intentionally positioned Harvey as a technology company that enables the legal profession, not one that competes with it.
“There is a reason the billable hour still exists,” Gabriel Pereyra said in a recent interview. “Legal work is unpredictable. But with better models, you can start to scope and price legal work more accurately.”
