The archive · AI & Models · Product decision · 2022–2025
Harvey bets domain-specific legal AI wins; $300M→$160M rounds to $8B in 2025
A lawyer and AI scientist's GPT-3 experiment becomes legal AI's $8B king, $100M ARR.
Harvey
What the business is
Harvey sells an AI platform that helps law firms and corporate legal teams research, analyze contracts, run due diligence, and draft documents using large language models tailored to legal work.
Starting capital：$5M from OpenAI Startup Fund (2022), then $21M Series A, $100M (Jul 2024), $300M Series D, $300M Series E, $160M (Dec 2025)
How it started
In 2022 Weinberg, a former securities and antitrust litigator, teamed with AI researcher Pereyra to test GPT-3 on legal questions pulled from Reddit. They sent the prototype to OpenAI's C-suite, landed a July 4 meeting, and walked out with OpenAI Startup Fund's first investment of $5M.
What happened
Harvey grew from 40 customers in 2024 to 235 in 42 countries, then past 250 enterprise customers in 43 countries including A&O Shearman, Verizon and PwC; it partnered with PwC to co-build the Tax AI Assistant. Series D of $300M at a $3B valuation came in February 2025 with ARR past $50M and an $8B-valued round of $160M led by a16z in December.
How it ended up
Harvey passed $100M annual recurring revenue in August 2025, counted 50 of the top AmLaw 100 firms among its customers, and closed a December 2025 round at an $8B valuation. It remained private and unprofitable disclosure-wise, still scaling internationally with a London office.
Background
Harvey began as a two-person experiment: Winston Weinberg, a first-year associate at O'Melveny, and Gabe Pereyra, a former DeepMind and Meta researcher, built a GPT-3 chatbot that answered legal questions and cold-emailed it to OpenAI. The prototype led to a July 4, 2022 meeting with OpenAI's C-suite and OpenAI Startup Fund's first investment, $5M.
The bet was that legal work — reading, summarizing, drafting, and reasoning over text — was a perfect fit for LLMs, but only if the product embedded deep legal context. Harvey hired lawyers from large firms and in-house teams to design and evaluate the product, turning a 'GPT-wrapper' into a domain-specific platform used for contracts, due diligence, litigation support, and compliance.
Traction followed: customers grew from 40 in 2024 to 235 in 42 countries, then past 250 enterprises including A&O Shearman, Verizon and PwC, with a PwC collaboration to co-build the Tax AI Assistant. Annual recurring revenue passed $50M by February 2025 and $100M by August 2025, with 50 of the top AmLaw 100 firms as customers.
The funding ladder matched the revenue curve: $300M Series D at a $3B valuation in February 2025, a $300M Series E at $5B in June, and a $160M round led by Andreessen Horowitz at an $8B valuation in December. Weinberg framed the shift as 'task displacement, not job displacement' — associates still do strategy while Harvey absorbs the busywork.
What has to be true
- A lawyer-founder who had lived the pain of first-year document grind gave the product credibility and domain fidelity competitors lacked.
- Cold-emailing a working prototype to OpenAI produced the single most valuable early check — OpenAI Startup Fund's first — plus a durable halo effect.
- Embedding lawyers in model design and evaluation created genuinely legal-grade output, defeating the 'GPT-wrapper' critique by becoming the layer on top.
- Landing top-of-market logos — A&O Shearman, PwC, Verizon and half the AmLaw 100 — made later enterprise deals self-reinforcing.
- The legal market's size (~$400B in the US) and text-heavy workflow justified valuations that looked extreme against ARR.
What can be applied
Domain expert plus model expert beats a generic model play: embedded lawyers made the product legal-grade, and the 'GPT-wrapper' dismissal became its moat.
Aftermath
As of December 2025 Harvey was scaling hard: $100M+ ARR, 50 AmLaw 100 customers, a London office for international growth, and an $8B valuation. It remained private with no IPO or exit announced; competition included Thomson Reuters' Casetext, Eve, Spellbook, Robin AI and LexisNexis-linked tools.
Sources
- Legal AI startup Harvey confirms $8B valuation
- Legal AI startup Harvey lands fresh $300 million in Sequoia-led round
- 32. Harvey — CNBC Disruptor 50
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card