Harvey Hits $15.5B—Its Open-Weight Strategy Fuels the Premium

SiliconANGLE’s September 9 report confirmed that legal AI company Harvey raised $550 million at a $15.5 billion valuation, with Diffusion and Lightspeed Venture Partners leading the financing. The deal increases Harvey’s private valuation by $4.5 billion from its previous round.

In its official funding announcement, Harvey connected the raise to Tenet, its first post-trained open-weight model, and Harvey LAB, its Legal Agent Benchmark; the company also said Harvey is used by 80% of Am Law 100 firms and five of the Fortune 10. The financing therefore backs a wider proposition than growth in a legal AI application: Harvey is moving into the models and evaluation infrastructure beneath its customer-facing software.

A $4.5 billion step-up since March

The latest valuation is approximately 40.9% above Harvey’s previous $11 billion mark. That percentage is an arithmetic comparison between two private financing valuations; it does not imply that revenue, profit or adoption increased at the same rate.

TechCrunch’s funding chronology places the new round after a $200 million financing at an $11 billion valuation in March and an $8 billion valuation in December, while putting Harvey’s cumulative funding above $1.55 billion. Based on those disclosed values, the company has added $7.5 billion to its valuation since December and has nearly doubled it over roughly nine months.

Harvey did not assign the September financing a conventional round label. More importantly, a private valuation reflects the price negotiated by the company and participating investors; unlike a public-market capitalization, it is not continuously repriced through open trading.

Tenet takes Harvey beyond the application layer

The strategic change is Harvey’s attempt to control more of the technical stack used for legal work. Its established products organize model capabilities into legal workflows, while Tenet moves the company into post-training a base model and modifying the agent harness that determines how the system executes extended assignments.

Harvey’s Tenet research preview identifies the system as a Kimi K3 base post-trained with Fireworks Research on synthetic, publicly available and human-expert data, without customer data. In Harvey’s internal evaluation, the complete Tenet configuration—including harness changes—completed almost twice as many held-out LAB tasks as base Kimi K3 and increased the all-pass rate by nine percentage points; on LAB Contracts, it completed 20% more tasks and raised that rate by two percentage points.

The published comparison measures the full configuration, including its execution harness, rather than the model weights in isolation. It is also a company-run benchmark result, not an independent assessment of performance on live client matters.

Tenet remains a research preview, with capabilities developed through the program expected to enter Harvey’s products over time. The available evidence establishes a technical direction, not a completed migration away from third-party proprietary models.

The larger ambition is to enable law firms to build specialized models they can control. If Harvey combines customer-specific expertise with its training environments and execution software, its role expands from presenting external models to supplying infrastructure through which legal intelligence can be adapted and operated.

Harvey LAB turns evaluation into infrastructure

Harvey LAB broadens the strategy by providing a common structure for testing long-horizon legal agents. A company developing both specialized systems and an evaluation framework can use comparable tasks to train models, refine their harnesses and identify where performance remains incomplete.

Harvey’s LAB documentation describes an open-source benchmark with more than 1,200 agent tasks across 24 legal practice areas and over 75,000 expert-written rubric criteria. Each task includes an instruction, materials for a synthetic client matter and a required legal work product; under the benchmark’s all-pass method, every criterion must pass for the task to count as complete.

Opening LAB to model providers, researchers, law firms and agent builders positions Harvey as a contributor to shared evaluation infrastructure as well as a software vendor. The benchmark can make differences among model-and-harness configurations more legible, but its scores alone cannot establish accuracy, productivity, reliability or financial returns in production.

The premium prices a broader but unproven thesis

The investment case now spans legal applications, post-trained models and evaluation infrastructure, with greater customer control over specialized intelligence. The round arrived as that strategy widened, but public information does not isolate how much of the $15.5 billion valuation investors assigned specifically to Tenet or LAB.

The adoption figures also require careful labeling. They are company-reported and come without published definitions of active use, deployment depth or independent auditing, so they indicate claimed market reach rather than independently verified engagement.

The financing included no detailed public metrics for profitability, cash flow, customer retention or the economics of model development. Outsiders therefore cannot separate the valuation effects of operating growth, investor demand, financing conditions and expectations for the open-weight program.

As of September 12, Harvey has secured financing at roughly a 41% premium to its March valuation while extending its strategy deeper into models and evaluation. What remains unknown is whether Tenet’s capabilities will reach production, whether customers will adopt controlled specialized models at meaningful scale, and whether independent evidence will validate the broader platform thesis.

Also read:

• Uzum’s $2.3 Billion Valuation Holds—Its Unicorn Milestone Came Earlier
• Apple to Open Apple Intelligence to Google and Anthropic Models in iOS 27
• Chinese Open-Weight Models Are Nipping at the Heels of Western SOTA