What it is
Harvey is the legal-AI assistant built for elite law firms and enterprise in-house teams. Backed by OpenAI as an early enterprise partner, it trained against legal-specific corpora and has the deepest commercial footprint in the AmLaw 100. By June 2026 Harvey passed $300M ARR — roughly 3x its $100M ARR of August 2025 — with 100,000+ lawyers across 1,300+ organizations. A March 2026 growth round co-led by GIC and Sequoia set its valuation at $11B. Used for contract review, legal research, document drafting, and matter management.
Why it shows up in Legal Ops stacks
- Built for legal workflows. Unlike generic LLMs, Harvey’s UI, retrieval, and prompting defaults assume legal context — case citations, jurisdictional awareness, contract clause libraries. Its Workflow Agents ship pre-built, no-code sequences for specific document types (M&A deals, lease agreements, credit agreements) that break a task into steps and pause for human review.
- Enterprise-grade governance. SSO, audit logs, matter-scoped access controls, ethical walls, and compliance certifications. The default choice for firms that can’t put client data into ChatGPT.
- Composable via MCP. Harvey now runs as both an MCP client — surfacing a firm’s internal tools and agents inside Harvey — and an MCP server, letting other systems call Harvey’s document analysis, vault review, and research, with centralized permissions applied across every connected system. More than 25,000 custom agents already run on the platform.
- Deep firm partnerships. Harvey co-designs workflows with major firms (A&O Shearman, PwC Legal). Features tend to ship aligned with how partner-track lawyers actually work.
A specific use case
M&A due diligence over a deal-room vault: point a Workflow Agent at hundreds of contracts, have it extract change-of-control, assignment, and MFN clauses into a review grid, and route only the exceptions to an associate — instead of a first-year reading every document from page one.
Pricing
Harvey is custom-quote only — no public pricing page, no free tier, and minimum commitments that typically start around 25 seats.
- Reported bands (third-party, triangulated — Harvey publishes none): mid-market in-house teams and smaller firms land near $1,000–$2,000 per seat/month; the Harvey Assistant baseline is reported around $12,000–$14,400 per seat/year.
- Enterprise volume discounts pull AmLaw-100 per-seat rates well below those figures, but full-firm deployments still run into seven figures annually.
- Customers report a 10–25% annual renewal uplift when the contract has no cap.
- Not viable for solo practitioners or small in-house teams; the alternatives below are the right path at smaller scale.
Best for
- AmLaw 100 / Magic Circle firms
- Enterprise in-house legal teams ($1B+ revenue, 20+ legal headcount)
- Firms whose primary blocker on AI adoption is governance/compliance, not capability
Alternatives and when to pick them instead
- Thomson Reuters CoCounsel — the other pole of the enterprise market. Pick it when Westlaw-grounded research fidelity and citation traceability matter more than workflow-suite breadth.
- Legora — the fastest-growing challenger; a 2026 mega-raise made it Harvey’s chief rival. Pick it for collaborative, European-origin drafting and review where teams want tighter in-document control.
- Spellbook — pick it for Word-native contract drafting at solo-to-midsize scale, where Harvey’s price floor and procurement cycle don’t pencil out.
Watch-outs
- Opaque, high pricing and a long procurement cycle. Guard: run seat-count-banded quotes from two vendors (e.g., Harvey and CoCounsel) in parallel, and negotiate a renewal-uplift cap into the first contract rather than the second.
- The legal-AI lead is narrowing. Harvey’s 2024–2025 head start is under pressure from CoCounsel, Legora, and Spellbook. Guard: scope a paid pilot against one named rival on your own matters and score accuracy on a fixed clause set before committing firm-wide.
- Data-residency and training-data policies vary by tier and region. Guard: get the DPA and residency terms in writing, and confirm client data is not used for training, before onboarding regulated matters.