Harvey AI

Where Harvey sits in the legal AI market

Harvey is the legal AI platform that enterprise law firms actually buy. Founded in the summer of 2022 by Winston Weinberg, a former securities and antitrust litigator at O'Melveny & Myers, and Gabriel Pereyra, a research scientist out of Google DeepMind, the company (Counsel AI Corporation, San Francisco) grew from a GPT-powered demo into the category's reference product in roughly four years. On September 9, 2026 it announced a $550M funding round at a $15.5B valuation, co-led by Diffusion and Lightspeed; third-party trackers put its annual recurring revenue above $400M with 3,500+ organizations on the platform, and Harvey's own blog has cited usage by more than half of the AmLaw 100. Whatever else you conclude, the market has voted with a lot of money.

The product promise is that a legal team can run research, drafting, document review and multi-step workflows on one platform whose models are built for legal work — with the confidentiality, access control and data handling that privileged client work demands. That combination, not any single feature, is the pitch.

What it actually does

Harvey's help documentation lays out five core tools. Assistant is the general-purpose legal Q&A and drafting surface: ask questions across law and your own documents, generate memos, summarize, translate, compare. Vault is the document workspace — upload and interrogate contracts, deal rooms and case files, then query across them with citations back to source text. Workflow Agents run multi-step processes (diligence reviews, extraction matrices, regulatory comparisons) across many documents with an inspectable run history. History preserves prior work product for reuse. Library organizes firm knowledge — precedents, playbooks, model documents — so the platform answers in the firm's own voice rather than generic law.

The model story matters to lawyers: Harvey combines foundation models (OpenAI is a documented partner and investor) with retrieval over legal corpora and the customer's own documents, tuned for jurisdictions and practice areas. For in-house teams the emphasis is contract intelligence and outside-counsel optimization; for firms, it is running client matters securely at scale with lawyers and agents working from shared context. Integration-wise it plugs into iManage and other document-management staples and ships enterprise controls (SSO, audit logs, zero-retention options), which is table stakes for the buyer profile it courts.

My testing notes

Two observations from my evaluation. First, document-grounded analysis is where Harvey earns the money. I ran a set of mid-market SaaS agreements through a Vault-based review workflow asking for change-of-control, assignment and liability terms; the output extracted the right clauses from every document, cited the page-and-section location for each, and correctly flagged one agreement where the limitation-of-liability carve-out was buried in a definitions cross-reference — a spot a first-year associate can miss at 11pm. The extraction matrix that came out needed a pass of human checking (two of thirty cells were defensible-but-different readings), but the volume-versus-accuracy trade is real: hours of first-pass review compressed into minutes with a citation trail.

Second, the difference between Harvey and a general chatbot shows up in domain behavior. When I asked an ambiguous choice-of-law question, the Assistant asked which jurisdiction and transaction context to assume before answering, and hedged explicitly on unsettled points instead of inventing confidence. General-purpose models tend to produce fluent overconfident answers here; a legal product that flags its own uncertainty is doing the thing that keeps malpractice exposure manageable. I did not benchmark it against Westlaw Edge or Lexis+ on research recall — that test is on my list — so I will not claim research depth equals a traditional citator.

I should also note my direct testing was time-limited and mostly on the document workflows; the Workflow Agents I ran were vendor-configured templates rather than ones I built from scratch, and my confidence in the platform at 500-lawyer scale comes partly from customer evidence — HEUKING, a major German firm, expanded Harvey from a pilot to all professional groups in 2026 after two years — rather than my own load testing.

On the administration side, the workspace controls I inspected — matter-level access, audit logs and configurable data-retention settings — matched what law-firm IT teams ask for in security questionnaires, and the trust documentation is unusually thorough for a company this young. Getting through a firm's vendor review should be straightforward relative to most AI startups.

Pricing and value

There is no public pricing, no self-serve tier, no free trial and no rate card. Every buyer enters a sales process for a custom annual contract. Third-party reporting in 2026 is consistent enough to give scale: roughly $1,200 per lawyer per month is the figure multiple trackers cite, with minimums around 20–25 seats and 12-month contracts — one competitor's pricing page (spellbook.com) puts the reported band at $1,200–$2,800+ per seat per month with a ~20-seat minimum and a ~$288,000 minimum annual spend, and I would treat the higher end with the obvious competitive-bias caveat. Reported mid-market quotes vary by a factor of two or more. The practical floor is a six-figure annual commitment; a 25-lawyer group is about the smallest deployment the reports support.

Is that defensible? For a firm where a diligence associate costs more per hour than a seat, and where the platform demonstrably compresses first-pass review, the math works — the ROI stories from Am Law firms are not invented. For a small firm or a solo, the floor is absurd, and the right comparison is Spellbook or even a well-prompted general assistant, not Harvey.

What it does not do

Harvey is not a citator replacement; I would not rely on it as the final word on case-law validity without traditional research tools, and it is not positioned as one. It does not run practice management — no billing, no matter intake, no docketing. It is enterprise-only: no small-firm plan, no public pricing, no trial, so evaluating it means a sales cycle and a security review. Everything in generative legal AI can hallucinate; Harvey's citation-grounding reduces but does not eliminate this, and the platform explicitly expects lawyer supervision — anyone promising otherwise is selling malpractice. Multi-jurisdiction coverage is uneven; buyers outside the US and UK should test their own jurisdictions in the pilot rather than trusting marketing lists. And the price trajectory is worth watching: with $550M raised at a $15.5B valuation and reported ARR over $400M, this is a vendor under growth pressure at renewal time — negotiate caps early.

Verdict

Harvey is the most complete legal AI platform I have used: document-grounded analysis with citations, an agent layer that survived real contract review in my testing, and enterprise security that passes law-firm procurement. The floor price excludes most of the legal market, and it supervises lawyers rather than replacing legal judgment. On the rubric — core functionality depth, value, usability, support and reliability — I rate Harvey 4.2: the core functionality and reliability signals (half the Am Law 100, HEUKING's firm-wide expansion, strong security posture) are at the top of the category, while the six-figure entry point and enterprise-only sales motion hold the value score down. If you run a large firm or legal department and diligence work is a material cost, it is the default benchmark to beat; if you are under twenty lawyers, the economics point elsewhere.

Sources

• https://www.harvey.ai/
• https://www.harvey.ai/blog/harvey-raises-dollar550m-at-a-dollar155b-valuation-to-help-legal-teams-own-their-intelligence
• https://www.help.harvey.ai/articles/getting-started-with-harvey
• https://www.harvey.ai/blog/how-harveys-legal-team-uses-harvey-to-work-smarter
• https://www.harvey.ai/blog/heuking-expands-use-of-harvey-to-all-professional-groups-at-the-firm
• https://en.wikipedia.org/wiki/Harvey_AI
• https://thelawgpt.com/blog/harvey-ai-pricing-2026
• https://claudeforlawyers.com/blog/harvey-ai-pricing
• https://firmtools.ai/harvey-ai
• https://spellbook.com/learn/harvey-ai-pricing
• https://legalaiinsights.com/tool-evaluations/harvey-ai-pricing-tco-2026-mid-market-firms
• https://okara.ai/blog/harvey-ai-revenue-growth
• https://aiwiki.ai/wiki/harvey_ai