Harvey Raises $550M and Trains Its First Legal Model on a Chinese Open-Weight System, Not Its Own Backer's Tech

Harvey, the legal AI company, has raised 550 million dollars at a 15.6 billion dollar valuation, a 41 percent jump from the 11 billion dollars it was worth just six months earlier, as the company shifts from reselling AI models built by others toward developing and owning its own.

The round was co‑led by Lightspeed Venture Partners and Diffusion, a newly formed firm co‑founded by longtime Harvey backer and former Coatue Management investor Kris Fredrickson. Sapphire Ventures and Whale Rock Capital Management joined as new investors, alongside a long list of returning backers including Sequoia Capital, Kleiner Perkins, Andreessen Horowitz, Coatue, Conviction, Elad Gil, Evantic, GIC, Goldman Sachs Alternatives, WndrCo, and Verified Capital. Harvey has now raised more than 1.5 billion dollars in total since its founding in 2022.

The company's valuation trajectory over the past 18 months has been extraordinary even by the standards of the current AI funding boom. Harvey was valued at 3 billion dollars in February 2025 following a Sequoia‑led round, reached 5 billion dollars four months later on a Kleiner Perkins and Coatue‑led raise, climbed to 8 billion dollars in December 2025 on an Andreessen Horowitz‑led round, and hit 11 billion dollars in March 2026 through a 200 million dollar raise co‑led by Sequoia and Singapore's GIC. This latest close puts Harvey's valuation more than five times higher than it was 18 months ago.

Harvey was founded in 2022 by chief executive Winston Weinberg, a former securities and antitrust litigator at O'Melveny & Myers, and president Gabe Pereyra, a former research scientist at Google DeepMind and Meta. The company builds an AI platform for law firms, in‑house legal teams, and professional services firms, covering contract analysis, compliance, due diligence, and litigation support. Rather than locking itself to a single foundation model provider, Harvey has built its platform to work across multiple underlying AI models and fine‑tune them specifically for legal work.

The most notable development accompanying this round is the introduction of Harvey Tenet, the company's first in‑house, post‑trained model. Tenet is built on Moonshot AI's open‑weight Kimi K3 model and was post‑trained using reinforcement learning with the help of inference provider Fireworks. The choice is a striking one for a company that OpenAI itself has previously backed and supplied models to: rather than building further on OpenAI's technology, Harvey instead chose a Chinese open‑weight foundation as the base for its first serious proprietary model. According to TechTimes, the strategic logic centers on data control. Harvey's bet is that responsibly handling privileged attorney‑client communications at scale requires owning the full model stack end to end, keeping law firm documents off third‑party servers entirely, a data sovereignty argument that becomes considerably harder to make credibly if the underlying model itself is licensed from an external lab with its own infrastructure and data handling practices.

Alongside Tenet, Harvey also launched Harvey LAB, its Legal Agent Benchmark, a tool intended to help law firms, in‑house teams, and professional services firms evaluate and benchmark AI performance on legal‑specific tasks as they build and post‑train their own models atop Harvey's infrastructure. Lightspeed partner Sebastian Duesterhoeft, who is joining Harvey's board as an observer, argued that in‑house corporate legal teams represent the larger prize within Harvey's addressable market, since they offer, in his words, "the direct path to go after all of legal services spend." Lightspeed reportedly views legal services as potentially the second‑largest addressable market for AI overall, trailing only software coding.

Harvey's commercial scale has grown quickly alongside its valuation. The company now serves more than 3,000 organizations, up from roughly 1,300 as of March, with annual recurring revenue that has crossed 400 million dollars. Roughly 80 percent of Am Law 100 firms use the platform, along with 20 percent of the Fortune 500 and five Fortune 10 companies, including named customers Latham & Watkins and Microsoft's in‑house legal team.

That commercial success has not gone unchallenged. Both of Harvey's own model suppliers have moved into direct competition with it: Anthropic has released legal‑specific plug‑ins for Claude, and OpenAI has partnered directly with law firms to customize ChatGPT for legal work, turning two companies Harvey depends on for underlying model access into rivals competing for the same customers. Closer to home, Swedish competitor Legora recently topped a 5.5 billion dollar valuation on a 600 million dollar Series D and, according to Tech Funding News, is separately in talks to raise at more than 10 billion dollars, up from 5.6 billion dollars in March, indicating the European legal AI market is scaling nearly as fast as Harvey itself.

Harvey has also been active on the acquisition front, buying AI agent security startup Guardrails AI as part of this funding cycle, marking its fourth acquisition of 2026. That string of acquisitions, combined with the shift toward proprietary models and Harvey's own benchmarking tools, suggests the company is working to build a defensible, vertically integrated position across the legal AI stack rather than remaining primarily a workflow layer sitting on top of models it doesn't control. The global legal AI software market is projected to grow substantially over the coming years, and with foundation model providers increasingly competing directly against the legal‑specific applications built on their own technology, Harvey's bet on owning its model stack end to end may prove less a strategic luxury than a competitive necessity as the boundary between infrastructure provider and application layer in legal AI continues to blur.