Harvey vs Casetext (CoCounsel)

This pairing has been dead on one side since April 1, 2025, when Thomson Reuters shut down Casetext. TR bought it in 2023 for $650M, started redirecting casetext.com on February 1, 2025, and retired the standalone platform two months after that. The CoCounsel skills survived and ship today inside Thomson Reuters CoCounsel, sold per attorney against Westlaw — the live platform bake-off is Harvey vs Thomson Reuters CoCounsel.

The reason to keep this page is that the argument behind the pairing has also collapsed, and that part is less obvious. Anyone comparing Harvey against Casetext was weighing one specific trade: frontier model quality on your own documents, versus grounded case-law research at a price a small firm could sign. Both halves have moved. Harvey now reaches primary law directly, and the cheap end of legal research AI is no longer $200 a seat. If you arrived here holding that trade in your head, it is the wrong trade.

The research gap that justified Casetext is closed

Casetext’s structural argument against Harvey was citation grounding: Harvey ran a strong model over your firm’s documents, but it could not tell you whether a case was still good law. That stopped being true in 2025.

LexisNexis and Harvey announced a strategic alliance in June 2025 that put LexisNexis primary law and Shepard’s Citations inside the Harvey platform. Ask LexisNexis went to open beta for all Harvey users on August 12, 2025: natural-language questions answered against US federal and state case law, statutes, administrative codes, and agency decisions, with pinpoint citations linking to full text and Shepard’s signals attached to check whether the authority still stands. The answers are generated by LexisNexis fine-tuned models running on LexisNexis infrastructure, not by Harvey’s model reading scraped case text.

That is the specific evidence that moves this verdict. It does not make Harvey a research database — it makes “Harvey has no verified primary law” a claim that was accurate in 2024 and is not accurate now. Note what it does not close: the equivalent grounding on the Westlaw corpus, with KeyCite, still belongs to CoCounsel, and a litigator whose citation habits are Westlaw-shaped is not being offered a substitute.

Where Harvey pulled ahead on everything else

• Agents, not fixed skills. Casetext shipped eight defined skills — document review, deposition prep, contract analysis, research memo, and four more — and that fixed surface was its adoption advantage with non-technical lawyers. Harvey launched roughly 500 prebuilt agents plus a self-service Agent Builder in May 2026, in early access rolling out over subsequent months. The Casetext model was pick from a menu; the Harvey model is build the menu.
• Published evaluation of its own agents. Harvey released the Legal Agent Benchmark on May 6, 2026 — open source, 1,200+ agent tasks across 24 practice areas, scored against roughly 75,000 expert-written rubric criteria, with Nvidia, OpenAI, Anthropic, Mistral, and DeepMind supporting it. Publishing a benchmark you can be beaten on is a stronger signal than a vendor accuracy number.
• Deployment weight. Harvey passed $300M ARR by June 2026, about 3x its August 2025 figure, across 100,000+ lawyers at 1,300+ organizations, at an $11B valuation set in a March 2026 round co-led by GIC and Sequoia. Casetext’s install base was migrated onto Thomson Reuters paper.

What Casetext was right about, and who inherits it

Casetext’s real product was not the model. It was the checkout. A solo attorney put in a card and was working the same afternoon, with no Westlaw contract, no procurement cycle, and no seat minimum. Nothing in this comparison inherits that.

Thomson Reuters publishes a small-firm number and stops there: Westlaw Advantage with CoCounsel Essentials runs $639.20/user/month for a solo attorney on All States & Federal coverage at a one-year term, with vendor-published term discounts of 12% at two years and 18% at three. Online pricing is available only to firms up to 10 attorneys; above that it is a sales cycle. So the closest thing to a Casetext successor costs roughly 3.2x what Casetext charged, and the self-serve ceiling is ten lawyers.

Pricing reality

Harvey has never published a price and still does not — the tool entry carries no starting figure because there is nothing to carry. Third-party reporting through 2026 clusters a base seat around $1,200/user/month for mid-market firms, in a band of roughly $1,000–$2,000+, with a LexisNexis-bundled seat estimated near $2,400/user/month and a reduced-scope tier reported near $399. Seat minimums are reported at 20–50 on 12-month terms, increasingly 2–3 year commitments, with 10–25% annual escalation where no cap is negotiated. Treat every one of those as directional; none is vendor-published, and the dispersion across sources is itself the finding.

Against Casetext’s roughly $200/user/month, that is a 6x multiple at Harvey’s reported base seat and about 12x for the Lexis-bundled seat — the configuration that actually replaces what Casetext did. The primary-law grounding described above is not free, and it is roughly the price of the entire Casetext subscription again.

Implementation effort

Casetext was same-day. Harvey is a procurement project: security review, document ingestion, tuning on firm precedent, and user training, with vendor-reported sales cycles around six months. CoCounsel sits between the two — days for a small firm buying Essentials online, longer for a full Westlaw-integrated deployment. If your evaluation window is a week, only one of these three can be evaluated in it, and it is not Harvey.

Verdict

Pick Harvey if you are an AmLaw firm or a large in-house team, your work is bespoke workflow automation and document analysis at volume as much as research, you want the platform tuned on your own precedent, and a six-figure annual commitment plus a multi-month runway is a budget line you already have. Buy the LexisNexis-bundled configuration if primary-law research is part of the job — the base seat is not the thing that replaces Casetext.

Do not pick Casetext. It cannot be bought. If you are pricing it from a review site or a procurement spreadsheet that still lists it live, that number is more than a year stale.

If you came for the Casetext price point, go sideways rather than up. Spellbook handles transactional drafting and review inside Word, Legora covers drafting and review without a tied research corpus, and Claude analyzes documents you supply. Guard: none of the three does primary-law research with citation validation, so anything headed for a filing still needs a Shepard’s or KeyCite check outside the tool.

If you cannot decide and Westlaw is already a line item, buy CoCounsel. You are paying once for a corpus you already own, and the model layer will commoditize faster than the content layer. If you pay for neither database and have fewer than ten attorneys, CoCounsel is still the pick, because it is the only option here with a bounded public number.

Two things to watch

• The Casetext brand outlived the product by more than a year. Comparison articles, review sites, and budget spreadsheets still carry a live $200 price against it. Guard: verify against thomsonreuters.com before Casetext reaches a shortlist or a renewal model.
• Harvey’s research grounding is Lexis-shaped, not Westlaw-shaped. A firm that standardized on KeyCite signals is changing citator, not just vendor, when it routes research through Harvey. Guard: run a sample of live matters through both citators before retiring the incumbent research seat, and keep the incumbent through one renewal cycle.