Specialized workflows and multi-model architecture
Harvey powers work across 80% of the Am Law 100. Winston Weinberg, a former O’Melveny & Myers associate, and Gabriel Pereyra, a researcher from Google DeepMind and Meta, founded the company in 2022. Sequoia led an $80 million funding round in 2023. The platform uses a multi-model stack drawing on OpenAI, Anthropic, and Google. While Harvey uses a multi-model stack combining OpenAI, Anthropic, and Google to route tasks to the best performing model, firms still report a 5 to 10 percent rate of citations that require manual verification. The platform includes Assistant for drafting and summarization alongside Vault for bulk document analysis. Vault allows a team to upload an entire data room to surface anomalies or missing provisions in minutes. Knowledge handles complex regulatory and tax questions by anchoring answers in trusted data sets. Agents execute multi-step tasks like research and drafting without manual prompting at every stage. Contract Intelligence analyzes portfolios to benchmark terms against market standards. The platform’s ability to handle multi-jurisdictional research and tax domains provides an advantage for global firms. One corporate lawyer reported saving 10 hours of work per week. The company reached a $15.5 billion valuation following a $550 million funding round in September 2026. The company’s founding follows a five-part evolution from a simple assistant to a massive infrastructure play. It began with broad assistant capabilities before moving through security, platform, and infrastructure stages. This progression allowed Harvey to build deep integration with Microsoft 365 products like Word, Outlook, and Teams.
Security controls and enterprise costs
Harvey builds on Microsoft Azure infrastructure to provide data isolation and encryption. The system maintains SOC 2 Type II, ISO 2701, and GDPR compliance to satisfy enterprise requirements. Deployment requires significant setup, including configuration, permissions review, and user training. You likely know that small firms cannot justify these minimums. For large organizations, the costs remain high because the software requires a significant financial commitment. Industry reports place per-seat costs between $1,000 and $1,200 per month. A 25-seat deployment with a 12-month term costs approximately $360,000 per year. The product targets Am Law 200 firms and well-funded corporate legal departments. Harvey uses Azure Kubernetes Service, Azure Database for PostgreSQL, and Azure Traffic Manager to manage data. The technology team can deploy models in different regions, such as Europe or Australia, to meet specific data residency requirements. The company also plans to expand its embedded legal engineering teams following its 2026 funding. In December 2025, the company launched Shared Spaces to let law firms and clients work in the same AI-powered workspace. Design partners like PwC and Gleiss Lutz use these features for M&A analysis and regulatory matters. Harvey uses Azure Blob Storage with bring your own key configuration to manage data.
| Feature | Estimated Cost | Target User |
|---|---|---|
| Per-seat license | $1,000 – $1,200/month | Large law firms |
| Minimum deployment | ~$360,000/year | Am Law 200 |
| Custom training | Included in enterprise | Large legal teams |
Comparing Harvey to research databases and LLMs
Harvey competes against integrated research platforms like Westlaw Edge AI and Lexis+ AI. Westlaw Edge AI bundles CoCounsel capabilities to offer natural language questions and case summaries. Lexis+ AI uses Shepard’s integration to check if cited cases remain good law and reports 93 to 96 percent accuracy on citation correctness. Harvey differs because it remains model-agnostic and focuses on firm-specific training. Smaller firms often find more value in using Claude directly for a fraction of the cost. A Claude Team seat costs roughly $20 to $25 per user per month. CoCounsel standalone also offers per-seat licenses for $200 to $300 per month. Harvey allows firms to train the model on internal templates and style guides to maintain consistency. This training helps the tool write like the firm rather than a generic AI. One report found AI tools perform 6 to 80 times faster than lawyers at document analysis. The 2025 Vals Legal AI Report noted Harvey Assistant among these high-speed tools. For contract-heavy practices, tools like LegalOn provide playbook-based review at scale. Will foundation model companies eventually move into the legal market and squeeze out specialized platforms?
