Harvey
Legal AI for law firms and in-house teams. Bulk document review in Vault, agentic drafting and research over licensed LexisNexis primary law, plus Agent Builder for firm workflows.
Overview
Harvey sells an application layer for legal work rather than a model. Vault stores and bulk-analyzes matter documents, Knowledge answers regulatory and tax questions against licensed sources, Spaces shares work securely across organizations and Agent Builder, added in March 2026, lets a firm package its own multi-step procedures without engineering help. The June 2025 alliance with LexisNexis gave exclusive generative access to US primary law, statutes and Shepard's Citations inside the product, which is the piece competitors cannot replicate on a short timeline.
The model story reversed direction in 2026. Harvey dropped its own fine-tuned legal model after frontier reasoning systems outscored it on BigLaw Bench, its internal evaluation, and now routes tasks across Anthropic, Google and OpenAI through a model selector. In May 2026 it open-sourced the Legal Agent Benchmark: over 1,200 tasks spanning 24 practice areas, scored against more than 75,000 expert-written rubric criteria, with contributions credited to Anthropic, OpenAI and Google DeepMind. That benchmark is worth reading even if the product never enters your building, because it shows how far agentic legal work still is from unattended.
Commercially it leads on distribution. The March 2026 round put $200M in at an $11B valuation, revenue passed a reported $350M annualized by mid-year and press reports in August 2026 describe a raise near $15.5B. Pricing stays unpublished, which is itself a signal about who the buyer is. Legora, valued at $5.6B and reportedly seeking more than $10B on $150M ARR, is the pressure point, and CoCounsel sits underneath both on price.
Key Features
- ✓ Vault for storing and bulk-analyzing matter documents across a whole deal or case file
- ✓ Knowledge research over exclusively licensed LexisNexis US primary law, statutes and Shepard's Citations
- ✓ Agent Builder, added March 2026, for packaging firm-specific multi-step procedures without writing code
- ✓ Model Selector that routes tasks to Anthropic, Google or OpenAI systems instead of one fixed backend
- ✓ Spaces for shared work across counterparties, with Memory carrying drafting preferences between sessions
- ✓ SOC 2 Type II, ISO 27001, ISO 27701 and ISO 42001 certification, which is what actually clears firm procurement
Where it holds
- • Distribution no rival matches: 2,400+ legal organizations, 200,000+ users and more than 75 of the Am Law 100
- • Exclusive generative access to LexisNexis primary law means citations resolve to real authority rather than model recall
- • Publishes its evaluations and open-sourced the Legal Agent Benchmark in May 2026, 1,200+ tasks across 24 practice areas graded on 75,000+ rubric criteria
- • Model-agnostic routing limits exposure to any single frontier lab
Where it breaks
- • Nothing is inspectable from outside. No self-serve tier, no trial worth the name and no public API
- • Retired its own fine-tuned legal model after frontier models beat it on BigLaw Bench, so the differentiation is now workflow, content and evals rather than modelling
- • The LexisNexis deal covers US primary law, so non-US matters fall back on general model reasoning despite the 70 country marketing
- • Price scales per lawyer with no usage floor, which punishes firms where adoption is uneven
My Take
A quiet admission this year tells you what you are actually buying: the in-house legal model was retired once frontier reasoning models beat it on Harvey's own BigLaw Bench. What a firm pays for now is permissions, citation plumbing into Lexis primary law and an evaluation harness, not a model edge. At a reported $1,200 per seat per month before the Lexis add-on, on a 20 seat minimum, that is a six figure line item to approve on faith. Legora at $5.6B is cheaper and closing on collaborative review, so the durable moat looks like trust and distribution.
Quick Info
- Pricing:
- paid
- Openness:
- Proprietary
- Starting at:
- No published price list. Third-party reporting through 2026 puts the base near $1,200 per seat per month, roughly $2,400 with the LexisNexis content bundled, with a smaller-firm tier around $399 and a 20 seat minimum on a 12 month contract. A credit-metered pay-as-you-go option exists. A 100 lawyer deployment is modelled at $1.97M to $2.25M in year one once implementation and training are counted. Treat all of these as leaks, not vendor numbers.
- Added:
- Aug 2026
- Updated:
- Aug 2026
Use Cases
Judge it on your own work
The notes above say where Harvey holds and where it breaks. The fastest check is your own workload.
Visit website ↗