Harvey has raised $550 million at a $15.5 billion valuation, and at this point the numbers almost distract from what’s actually interesting about the company. According to TechCrunch, the round was co-led by Diffusion and Lightspeed Venture Partners, and it comes just six months after Harvey raised $200 million at an $11 billion valuation. Before that, in December, it was valued at $8 billion. The pace here is not normal.
Total capital raised now sits above $1.55 billion. Harvey has completed at least eight priced rounds since 2023, five of them since 2025, according to Pitchbook estimates. Some were extension rounds, so this one likely qualifies as a Series F in practical terms, though Harvey isn’t labeling it. That kind of naming ambiguity is a tell. When Databricks does it, it’s because the round doesn’t fit neatly into a traditional growth narrative. Harvey seems to be in similar territory.
The funding matters, but the model strategy matters more. A couple of weeks before this raise, Harvey released Harvey Tenet, its first in-house model. It was built on top of Kimi K3, an open-weight model, and post-trained on legal data with help from Fireworks, the inference provider that also helped Cursor build its own model. That’s a meaningful detail. Harvey is not building on GPT-4 or Claude and hoping for the best. It is actively moving away from dependency on OpenAI and Anthropic.
And it’s pushing that same thinking onto its customers. Harvey is encouraging law firms and legal teams to adopt and post-train their own open-weight models rather than defaulting to proprietary frontier APIs. That’s a different kind of product vision. It positions Harvey less as a chatbot layer on top of someone else’s model and more as the infrastructure through which an entire profession manages its AI adoption.
This matters because the legal sector has historically been one of the most cautious industries around data privacy, liability, and vendor lock-in. The pitch that firms can own and fine-tune their own models, rather than sending sensitive client data to a third-party API, is exactly the right argument in that context. Harvey is essentially telling law firms: you don’t have to trust OpenAI, you just have to trust us to help you build your own stack.
Competitors like Casetext, acquired by Thomson Reuters, and tools built on top of standard LLM APIs are now competing against a company with proprietary training infrastructure, a purpose-built model, and $1.55 billion in the bank. That gap is getting harder to close.
