Harvey has been in fundraising mode so relentless that its CEO, Winston Weinberg, recently told Business Insider he has not touched the majority of rounds the company has raised. On Wednesday, the legal AI company announced another $200 million, this one co-led by GIC and Sequoia, valuing the business at $11 billion — roughly 3.5 times the valuation it carried thirteen months earlier, when Sequoia first backed it at $3 billion. The trajectory has been San Francisco-implausible: $5 billion last June, $8 billion in December, and now this.
The comparison Harvey does not want you to make is to OpenAI. Weinberg has made a point of saying the $11 billion is a business valuation, not a model lab valuation. That distinction matters, because the thing that determines whether a vertical AI company is worth $11 billion is not how impressive its demos are but whether its customers actually pay it like a real business. By that measure, the numbers are catching up to the narrative. Harvey says it hit $190 million in annual recurring revenue in January, up from $100 million last August — a near-doubling in five months. Sacra, the research firm that tracks ARR independently, estimates $195 million by year-end 2025, up 3.9x from $50 million the prior year. The company says annualized revenue is now significantly north of $200 million. At those numbers, an $11 billion valuation is not absurd on a revenue multiple — it is aggressive but not incoherent.
More than 100,000 lawyers across 1,300 organizations now run work through Harvey, the company says, executing M&A due diligence, contract drafting, and document review across the majority of the AmLaw 100. More than 25,000 custom agents operate on the platform. Harvey acquired Hexus in January — a San Francisco company that builds AI tools for creating product demos and videos — to accelerate its product motion for in-house legal teams. It hired a chief product officer in February and, earlier this week, a chief strategy officer from Gibson Dunn. These are not the moves of a company conserving capital.
The thing Weinberg said to Business Insider that landed hardest was this: The things that we wanted to build over the next three years, we can probably build in one now. That is either a significant update to the companys trajectory or an unusually candid admission that foundation models have compressed what Harvey thought was a multi-year roadmap. The second interpretation is the interesting one.
Harvey originally trained a proprietary legal model — a vertical fine-tune built on legal text. In a post published last year, the company said its proprietary models substantially outperformed publicly available LLMs on its own BigLaw Bench evaluation. Since then, general foundation models have improved at baseline legal reasoning. Harvey incorporated Anthropic and Google models alongside OpenAI, optimizing its own systems on top of whichever model works best for a given task. Sacra, the research firm, frames this as Harvey abandoning its proprietary approach after frontier reasoning models from Google, xAI, OpenAI, and Anthropic began outperforming its custom model on BigLaw Bench — a characterization Harvey does not use in its own post describing the change.
Sequoias Pat Grady framed it this way to CNBC: Harvey sort of wrote the playbook for what it means to be an AI-native application company, which is the same thing Salesforce did back in the day with the cloud transition. That is a generous parallel. Salesforce used the cloud to unbundle enterprise software; Harvey is betting that agents — not copilots, not chatbots, but autonomous workflows executing legal tasks end-to-end — are the unit of its next chapter. The pricing is harder to pin down. Harvey does not publish a price list, a common practice for enterprise software sold to large law firms. Market estimates, including from Sacra, put Harveys base offering at roughly $1,200 per lawyer per month with twelve-month commitments and minimums around twenty seats. This is BigLaw math. A firm putting two hundred lawyers on the platform is writing Harvey a $2.9 million annual check.
The competitive argument is where Weinberg gets prickly. Harveys public position is that it competes with OpenAI and Anthropic for the serious end of the legal AI market, not with Legora or other legal-tech startups. Legoras chairman, Benchmarks Chetan Puttagunta, offered a different read to Business Insider: The lady doth protest too much, methinks. Legora counts roughly 20 percent of the AmLaw 100 as customers. Harvey says it has more than half. Legora does not disclose revenue. Harvey does. Whether that asymmetry reflects winners confidence or a need to justify a valuation built in fourteen months depends on who you ask.
Weinbergs sharper concern is not Legora. I think any company right now, the worst mistake you can possibly make is to become complacent, because how you build a company is completely changing, he told CNBC. He sees the model labs themselves as the more credible threat — OpenAI and Anthropic deciding that legal workflow is a product surface worth owning directly, not just an API call routed through a vertical. It is a reasonable fear. When Anthropic launched a legal plugin for Claude Cowork in early February 2026, investors fled legal-tech stocks: Thomson Reuters fell 19 percent, RELX fell 15 percent, LegalZoom fell 18 percent, Wolters Kluwer fell 13 percent. The market reaction was visceral and immediate, and it was not triggered by a legal-tech startup. It was triggered by a model lab shipping a feature.
That episode is the backdrop for why Harvey is raising like it is running out of time. With more than $1 billion raised across four disclosed rounds since February 2025, the company is stacking capital while the competitive window is open. The model labs are still APIs. The law firms are still locked in multi-year procurement cycles. And Harvey has 100,000 lawyers who have already built workflows on its infrastructure. Whether that moat is defensible against a model lab with a legal product team and a direct sales force is the question that $11 billion is betting the answer to.
