AI Agents for Lawyers

Comparisons

Harvey vs CoCounsel: Platform Play vs Westlaw-Grounded Research

Harvey vs CoCounsel for 2026: Westlaw and KeyCite grounding against firm-wide agents, published prices from $104 versus a 20-seat minimum, and a scenario table.

By Dr. Niklas SchmidtUpdated 13 November 20268 min read
On this page
  1. Two products that grew towards each other
  2. Grounding versus breadth
  3. What each costs, and who will tell you
  4. The Claude connection both share
  5. Agents and workflows compared
  6. Small-firm fit
  7. Scenario table
  8. A grounding test to run before you buy
  9. Running both, and running neither

Thomson Reuters will tell you what CoCounsel costs; Harvey will not. That asymmetry summarises the comparison. CoCounsel Legal is a Westlaw-grounded research and drafting product with published tiers from $104 a month, now rebuilt as agents. Harvey is a $15.5bn agent platform with a reported 20-seat minimum that connects to research rather than owning it. Choose by where your hours go: research and briefs, CoCounsel; transactions and cross-practice workflows, Harvey.

Two products that grew towards each other

CoCounsel started life as Casetext's research assistant and became Thomson Reuters' flagship after the 2023 acquisition. The version that matters now is the "next generation of CoCounsel Legal", generally available in the US on 20 August 2026, built on Anthropic's Claude Agent SDK plus Thomson Reuters' own legal model. It adds Westlaw Brief Builder, Tabular Analysis (10,000 documents against 100 questions in a pass), Deep Research Verify, matter Workspaces and a Word drafting agent. Thomson Reuters says a million professionals in 107 countries use CoCounsel, with Canada, the UK and Australia due later in 2026, and reports a Net Promoter Score of 77.

Harvey started as a wrapper on OpenAI models for Allen & Overy in 2022 and is now a platform: Assistant, Vault, Knowledge, Workflow Agents, Agent Builder, Memory and matter-aware Spaces since Harvey II (18 August 2026), 500-plus pre-built agents, 25,000-plus customer-built agents and 700,000-plus agent tasks a day. It raised $550m at $15.5bn on 9 September 2026 with ARR above $400m and claims 80% of the top 100 law firms.

The ILTA 2026 Technology Survey (14 September 2026) has them nearly level in adoption: CoCounsel at 44% of firms, Harvey at 43%. Among firms with 700-plus lawyers, Harvey reaches 84%. The reviews of CoCounsel Legal and Harvey go into the products individually.

Grounding versus breadth

The core architectural difference is where the authority comes from.

CoCounsel answers from Westlaw's editorial corpus, applies KeyCite treatment flags, and now runs a Deep Research Verify step that checks what it cites. That matters because of the numbers. Stanford RegLab's 2024 study found Westlaw AI-Assisted Research hallucinating 33% of the time and Lexis+ AI 17%; the tools tested were the previous generation, and Thomson Reuters has not submitted the current one to a comparable public test. A LexisNexis survey of 543 professionals published on 1 September 2026 found 81% feel more comfortable with AI grounded in legal sources, up from 70% a year earlier. Thomson Reuters' CTO Joel Hron frames the whole market around that preference: "The control point isn't where work starts. It's whether the output is accurate, grounded in authoritative sources."

Harvey's answer is breadth. Its agents draw on your documents (Vault), your firm's knowledge (Knowledge), whatever research connectors you plug in, and the open web. On the Vals Legal AI Report of February 2025, Harvey scored 94.8% on document Q&A against a lawyer baseline of 70.1%, but lawyers beat it at redlining (79.7% to 65.0%). Harvey did not appear in Vals' October 2025 research study; nor did Westlaw or Lexis, which declined to take part.

The trap is to believe grounding removes the verification duty. It does not. In Lacey v. State Farm (C.D. Cal., 6 May 2025) the court imposed $31,100 in sanctions on Ellis George and K&L Gates for nine bad citations out of 27; the tools involved were CoCounsel, Westlaw Precision and Gemini. Fletcher v. Experian (5th Cir., Feb 2026) reportedly traced 16 fabricated quotations to paid legal tools too. Grounded tools fail less often; they still fail. The verification protocol is the same whichever platform produced the draft.

What each costs, and who will tell you

CoCounsel LegalHarvey
Price transparencyPublished range about $104–$639 per user per month; solo Westlaw Advantage plus CoCounsel Essentials $639.20Not published; secondary estimates ~$1,000–1,200 per seat per month plus consumption
Seat minimumNone stated; solo bundles existReported 20 seats
BundlingSold with Westlaw and Practical Law tiersStandalone platform; research via connectors
Cost for a 10-lawyer firm (indicative)$12,500–$77,000 a year depending on tierUsually not available below the minimum
Cost for a 50-lawyer firm (indicative)$62,000–$383,000 a year$600,000–720,000 a year before consumption, at reported rates

The published CoCounsel range comes via GC AI's pricing roundup; Harvey's estimate comes from secondary sources. Neither number will survive a negotiation unchanged, and Harvey's contracts are reported to be adding token-based components. The Harvey pricing page and the pricing guide place both against Legora, Lexis+ Protégé and Clio Work at $199.

If you already pay for Westlaw, CoCounsel's marginal cost is the difference between your current bundle and one that includes it, which is often the cheapest legal AI a firm can buy. That is the single strongest argument for CoCounsel in a mid-sized US practice.

The Claude connection both share

Since August 2026 the two products share a supplier. Thomson Reuters rebuilt CoCounsel Legal on the Claude Agent SDK, the same primitives (bash, read, write, edit, sub-agents, MCP servers, allowed tools and permission modes) that power Claude Code. Harvey turned on Claude Opus 5 on 24 July 2026 and Fable 5.1 on 1 September, alongside GPT-5.6 Sol. Both also appear inside Anthropic's own offering: Claude for Legal ships a partner plugin, cocounsel-legal, that runs Westlaw Deep Research for users with a CoCounsel subscription, and lists Harvey among its 20-plus MCP connectors.

Two consequences. First, model quality is no longer a reason to choose one over the other; what differs is the content, the workflows and the governance around the same models. Second, a firm that already runs Claude Team or Enterprise can reach CoCounsel's research through the Claude for Legal plugins without buying a second interface, and can reach Harvey the same way. The vendors would rather you stayed inside their own product, a dynamic Artificial Lawyer calls the "battle for centrality".

Agents and workflows compared

Capability (Sep 2026)CoCounsel LegalHarvey
Pre-built agentsBrief Builder, Deep Research, Drafting Agent in Word, Tabular Analysis, Workspaces500+ practice agents; Agent Builder (early access May 2026)
Custom agentsLimited; workflows within the product25,000+ built by customers; agents "flag moments requiring user confirmation"
Document review at scaleTabular Analysis 10,000 documents × 100 questionsVault; diligence agents (GSK Stockmann: 15–20% saved on standard diligence, up to 75% on unstructured data rooms)
Research groundingWestlaw, KeyCite, Practical Law; Deep Research VerifyVia connectors (CoCounsel, LexisNexis, vLex, CourtListener)
WordDrafting agent inside WordWord and Outlook add-ins
Evidence and e-discoveryReveal and Everlaw integrationsEverlaw integration due autumn 2026
Ethical walls / DMSiManage and NetDocuments MCPIntapp ethical-wall enforcement (23 Jul 2026); iManage, NetDocuments MCP
Own modelTR legal LLM alongside ClaudeHarvey Tenet (open-weight, research preview)
Benchmark it publishesNone currentBigLaw Bench; LAB (open source, no leaderboard)

The pattern: CoCounsel's agents are few and deep and sit on the research corpus; Harvey's are many and configurable and sit on your documents. A litigator writing a summary-judgment brief on Ninth Circuit law wants the former. An M&A team running the same diligence grid across 40 data rooms a year wants the latter. The legal research use case and the due-diligence use case show both workflows end to end.

Small-firm fit

Below 20 lawyers, Harvey's reported minimum decides the question, and Harvey's CEO has said the company is moving downmarket only gradually. CoCounsel has solo bundles at published prices, Thomson Reuters gave CoCounsel free to 200-plus law schools in January 2026, and Clio Work, built on vLex's Vincent with a billion-plus documents across 110-plus jurisdictions, is $199 per user per month standalone since 21 April 2026. The small-firm guide and is Harvey worth it for a small firm go through the alternatives. A ten-lawyer US litigation practice that already holds Westlaw should add CoCounsel first and revisit Harvey when the minimum falls.

Scenario table

Your situationPickWhy
US litigation firm, Westlaw subscriber, 5–50 lawyersCoCounsel LegalGrounded research, KeyCite, Brief Builder, marginal cost on an existing bundle
Transactional firm, 50+ lawyers, diligence-heavyHarveyVault, diligence agents, Agent Builder, client alignment
Full-service firm, 200+ lawyersBothResearch in CoCounsel, workflows in Harvey; both reachable through Claude for Legal
In-house team with a Westlaw or Practical Law contractCoCounsel LegalPractical Law grounding; Word drafting agent
In-house team whose outside counsel share Harvey workspacesHarveyShared matters; 500+ in-house teams already on it
UK, Canadian or Australian firmHarvey now, CoCounsel when it ships locallyNext-gen CoCounsel is US-only at GA; UK launch of the prior version was 26 Jan 2026
Firm already on Claude EnterpriseStart with the pluginscocounsel-legal plugin and Harvey connector; buy the platform once usage proves it

A grounding test to run before you buy

Ask both vendors for a pilot and run the same five questions in each, plus in a plain Claude or ChatGPT business account as a control. Write the answers down before you look.

1. A question on a point your team argued last year, where you know the
   leading authority and one adverse case. Score: did it find both, and
   did it flag the adverse case's treatment?
2. A question that is settled in one state or circuit and open in
   another. Score: did it notice the split, or answer as if the law were
   uniform? (Every tool in Vals' 2025 study lost 11–14 points on
   multi-jurisdictional questions.)
3. A question with a false premise ("Given that the 2024 amendment
   abolished X, ..."), where no such amendment exists. Score: did it
   correct you or build on the error?
4. A request for three quotations from a named opinion you have in front
   of you. Score: are the quotations verbatim?
5. A request to draft one section of a brief from the verified
   authorities only, citing nothing else. Score: did it add authorities
   anyway?

Score each tool out of ten, then open every citation in the underlying source. Session 2 of the AI Lab for Lawyers course runs exactly this exercise on research tools, and the results usually surprise the people who assumed grounding meant finished.

Running both, and running neither

Large firms increasingly run both products and route work between them, which Law.com has called Big Law's "AI Tower of Babel". That is expensive but defensible: the research corpus and the workflow platform are different assets. Below the enterprise tier, the decision is simpler. Buy the product that grounds the work you actually bill, keep a $25 business chat tool for everything else, and remember the number that governs both choices: Charlotin's database logged 2,041 hallucination decisions worldwide by 14 September 2026, 811 of them involving lawyers, and the sanctions cases include firms with every tool on this page.

Frequently asked questions

Is CoCounsel better than Harvey for legal research?

For US legal research, yes, as of September 2026. CoCounsel Legal answers from Westlaw content with KeyCite treatment and a Deep Research Verify step; Harvey is a workflow and agent platform that connects to research databases rather than owning one. Harvey is stronger for transactional agents, diligence grids and firm-wide custom workflows. Many large firms run both.

How much does CoCounsel cost compared with Harvey?

Thomson Reuters publishes a range of about $104 to $639 per user per month depending on bundle; a solo Westlaw Advantage plus CoCounsel Essentials package lists at $639.20. Harvey publishes nothing; secondary sources estimate $1,000 to $1,200 per seat per month with a 20-seat minimum plus consumption charges. For a ten-lawyer firm, CoCounsel is typically the only one of the two that can be bought at all.

Can I use CoCounsel without a Westlaw subscription?

CoCounsel Legal is sold in bundles that pair it with Westlaw content, and its grounding depends on that content, so in practice you buy or already hold a Westlaw or Practical Law subscription. Thomson Reuters lists tiers from CoCounsel Essentials upwards; check which Westlaw tier and jurisdictions your bundle includes, since Canada, the UK and Australia follow the US release later in 2026.

Do Harvey and CoCounsel both use Claude?

Yes. Thomson Reuters rebuilt CoCounsel Legal on Anthropic's Claude Agent SDK plus its own legal model, with general availability in the US on 20 August 2026. Harvey runs a multi-model stack across OpenAI, Anthropic, Google and Mistral, added Claude Opus 5 on 24 July 2026 and Fable 5.1 on 1 September 2026, and released its own open-weight model, Tenet, in research preview. CoCounsel is also a partner connector inside Claude for Legal.

Put this into practice

AI Lab for Lawyers: a live, hands-on course for practising lawyers

4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.

Reserve a seat · €200 Maven money-back guarantee · certificate of completion