Harvey Raises $550M Series H at a $15.5B Valuation

A law firm can buy access to the same frontier models as its competitors. It cannot buy its own accumulated judgment off the shelf. Harvey is financing the harder project: turning firm knowledge, standards and workflows into intelligence that legal organizations can train, test and govern.

The company announced a $550M financing at a $15.5B valuation on September 9, 2026. Diffusion and Lightspeed co-led the round, which Lightspeed describes as Series H. Harvey says the capital will fund people, compute, specialized products and the model work behind its effort to help legal teams own more of the intelligence produced inside their organizations.

What Harvey Raised and Who Backed It

Harvey's investor list reflects how aggressively capital has moved toward legal AI. Returning participants included Sequoia, Kleiner Perkins, a16z, Coatue, Conviction, Elad Gil, Evantic, GIC, Goldman Sachs Alternatives, Verified Capital and WNDR. Diffusion, Lightspeed, Sapphire Ventures and Whale Rock joined as new investors, according to Harvey.

The company did not give the round a series label in its own announcement, so Series H should be read as the designation used by Lightspeed, one of the co-leads. Harvey also did not disclose the primary and secondary mix, ownership sold, share price or whether the $15.5B figure is being presented on a pre-money or post-money basis. Those details matter when a private-company valuation begins carrying almost as much attention as the product.

The financing follows an unusually fast climb. Harvey raised $160M at an $8B valuation in December 2025, then raised $200M at an $11B valuation in March 2026. TechCrunch reported that the latest round pushes Harvey's total funding above $1.55B. The capital history establishes the speed; it does not remove the need to distinguish funding, valuation and operating performance.

Why Owning Intelligence Changes the Product

Harvey's phrase “own their intelligence” addresses a practical enterprise problem. Legal work is shaped by precedent, review comments, negotiation positions, client histories, matter structures, permissions and professional judgment. A general-purpose model can provide broad capability, but it does not arrive knowing how one law firm evaluates a clause, how an in-house team allocates risk or which work product a client considers acceptable.

That is why Harvey is expanding beyond a conventional software layer. Harvey Tenet, the company's first post-trained open-weight model, uses a Kimi K3 base and was trained with Fireworks for long-horizon legal work. Harvey says the early research improved performance on held-out Legal Agent Benchmark tasks while preserving cost efficiency. The company presents those findings as initial research results, not as proof that one model performs uniformly across every legal workflow.

Harvey LAB gives the company an evaluation environment for realistic legal tasks, while its agent infrastructure routes work across models, tools and large document collections. Harvey's engineering team has argued that multi-model access is becoming necessary for quality, cost, conflicts and resilience. The product thesis is therefore broader than generating text. Harvey wants to decide which intelligence should handle a task, measure whether it completed the work correctly and preserve the controls required when the work belongs to a client.

Distribution Meets Model Economics

Harvey enters that technical expansion with unusually deep distribution. The company says it has crossed $400M in annual recurring revenue and serves more than 3,000 customers, including 80% of the Am Law 100, 20% of the Fortune 500 and five of the Fortune 10. Those figures come from Harvey and CEO Winston Weinberg rather than an independent audit, but they clarify the market assumption behind the valuation: investors expect legal AI adoption to move from individual assistance into organization-wide infrastructure.

Distribution creates data about workflows, failure modes and buying behavior, yet it also increases the burden. Legal agents must respect access controls, preserve citations, navigate long matters, apply firm-specific standards and behave predictably when the source material is incomplete. Every additional workflow can create more model calls, more evaluation work and more ways for a small error to become a client problem.

Compute economics now sit beside legal accuracy. A long-running agent may call models and tools hundreds of times across a large matter, making a “best model for everything” strategy expensive and operationally fragile. Harvey's investment in open-weight post-training, routing and task-specific evaluation is an attempt to control that curve without asking legal teams to accept lower standards for consequential work.

People and Compute Are the Operating Plan

Co-founders Winston Weinberg and Gabe Pereyra said the financing will support the next wave of AI transformation for law firms, in-house legal teams and professional-services organizations. Weinberg later described the plan more directly: hire across product, engineering, research and go-to-market, expand specialized products for contracting, litigation, deals and compliance, and invest in the compute and data behind the models that follow Tenet.

That combines two different scaling problems. Harvey must recruit people who understand frontier AI, enterprise software and legal work while building infrastructure that can train and run specialized systems economically. CTO Siva Gurumurthy's engineering organization is responsible for making those systems reliable across a growing global footprint, while Pereyra's research background connects product distribution with model and evaluation work.

Harvey's separate acquisition of Guardrails AI on September 9 adds another part of that plan. Guardrails built simulation and control systems for agent behavior, and its team is joining Harvey's product and engineering organization. The acquisition is not part of the $550M financing, but its timing shows where Harvey expects the difficult work to move: from getting an agent to produce an answer toward knowing how it behaves before a client discovers the failure.

What the $15.5B Valuation Asks Harvey to Prove

Harvey's valuation has nearly doubled since December 2025, while its strategy has become more technically demanding. The company is competing with specialist legal-AI vendors, legal-information incumbents and frontier-model providers that can move directly into professional workflows. It must also persuade customers that deeper model ownership will improve capability and control without weakening the flexibility of a multi-model platform.

The strongest evidence is not the size of the round. It is whether Harvey can turn its position inside legal institutions into better products, more reliable agents and lower-cost execution while preserving the trust that won those customers. Tenet, LAB, cloud agents and Guardrails AI give Harvey more of the stack to work with. They also give the company more of the outcome to own.

Legal organizations will keep accumulating judgment in documents, workflows and people whether Harvey captures it or not. The $550M wager is that Harvey can make that intelligence usable without stripping away the permissions, provenance and professional responsibility that made it valuable in the first place.