Why Harvey Built Its Own Legal AI Model Before Raising 550 Million Dollars
Legal AI startup Harvey has raised $550 million at a $15.5 billion valuation co-led by Diffusion and Lightspeed, landing days after it released Tenet, its first in-house model built on Moonshot AI's open-weight Kimi K3 rather than a partner's proprietary system.
Harvey just made a move that says more than its valuation does. The legal AI company raised $550 million at a $15.5 billion valuation, co-led by Lightspeed Venture Partners and newcomer Diffusion — but the detail worth pausing on is that the round landed days after Harvey shipped Tenet, its first in-house model, built not on OpenAI's technology but on Moonshot AI's open-weight Kimi K3.
OpenAI has backed Harvey since 2022 and remains an investor, which makes the base-model choice a deliberate hedge rather than an accident. The business numbers underneath the raise explain the confidence:
- Annual recurring revenue has passed $400 million, with more than 3,000 organizations on the platform
- Customers include 80% of Am Law 100 firms, 20% of the Fortune 500, and in-house teams at companies like Microsoft
- Valuation climbed from $11 billion in March, when the customer count sat at 1,300 — roughly $4.5 billion in growth in six months
CEO Winston Weinberg framed the strategy bluntly, saying all software companies need to turn into AI companies and that post-training models is part of that process.
The practical argument is about privilege: by downloading open weights and running training and inference on its own infrastructure, Harvey keeps law firm documents off third-party servers entirely.
Harvey also acquired Guardrails AI alongside the round, a security platform that stress-tests agent behavior — its fourth acquisition this year, and a signal that as legal agents take on longer autonomous tasks, verifying what they do is becoming as commercially valuable as making them capable in the first place.
