Skip to content
Published on

AI Legal Tech 2026 Deep Dive — Harvey, CoCounsel, Spellbook, Robin AI, Casetext, vLex Vincent, Relativity aiR

🇰🇷KO🇺🇸EN🇯🇵JA
Split View필사 모드
Authors

Opening — The slope of legal tech in May 2026

Law is, at its core, a text industry. The average lawyer processes more words per day than almost any other professional. So when ChatGPT shipped in November 2022, the first profession to shake was not coders — it was lawyers. By May 2026, that shaking has become tooling.

Recent surveys of AmLaw 100 firms (the top 100 US firms by revenue) show that more than 80 percent are piloting AI in some form, and 30 to 40 percent are already running it in production. All five UK Magic Circle firms have either adopted Harvey or built their own AI. Three of the four big Tokyo firms use AI review tools. In Korea, Kim and Chang, Lee and Ko, and Bae Kim and Lee are all evaluating in-house LLMs or external tools.

This guide maps the major legal tech AI tools by category as of May 2026, with real pricing, strengths, weaknesses, real incidents like hallucinated citations, the local Korean and Japanese ecosystems, open-source alternatives, and the 12 to 24 month forecast. The aim is one map that lawyers, legal ops managers, and legal tech builders can all use.

A category map — where AI entered the lawyer workflow

Legal tech AI splits into seven categories. Each has different leaders, different model dependencies, and different hallucination exposure.

  1. Contract drafting — first-pass NDA, MSA, SOW. Spellbook, Genie AI, Definely, CoCounsel Drafting.
  2. Contract review — clause flagging, risk assessment. Robin AI, Luminance, Kira, Diligen, Eve AI.
  3. Contract lifecycle management (CLM) — signing, renewal, expiry tracking. Ironclad, LinkSquares, Lexion (Docusign), Evisort.
  4. Legal research — case law search, citation verification. Westlaw Precision, Lexis+ AI, vLex Vincent, Casetext (now CoCounsel), Paxton AI.
  5. E-discovery — large-scale document review. Relativity aiR, Everlaw, DISCO, Logikcull, Reveal.
  6. M&A due diligence — data room analysis. Kira Systems, Luminance, Hebbia, Diligen.
  7. Memo and general assistant — Harvey, CoCounsel, Lexis+ AI, in-house chatbots.

Each category looks like an independent market, but by 2026 the strategic split between platforms (Harvey aiming at multiple categories) and verticals (Spellbook digging deep into Word-based drafting) has become very clear.

Harvey AI — the platform elite firms chose first

Harvey started in 2022 with OpenAI Startup Fund investment. Through Series C and D in 2024 and 2025, valuation pushed past three billion dollars, and 2026 revenue is reported to have crossed one hundred million dollars. Its strength is its partnership roster.

Headline partners:

  • Allen and Overy (now A&O Shearman) — global deployment across roughly 4,000 lawyers.
  • PwC — global license for around 4,000 legal professionals.
  • Clifford Chance — UK Magic Circle, global deployment.
  • Bain and Company — consulting, but using Harvey for contracts and research.
  • Others — Latham, Paul Weiss, Cleary, Macfarlanes, and more.

The core feature set is case research (US, UK, EU law), contract analysis and comparison, memo and email drafting, multilingual translation, and Workflows — a feature that lets firms wire their own pipelines on top of LLM steps.

Pricing is not public, but AmLaw firms typically contract in 100 to 1,000 seat blocks at enterprise rates believed to exceed 5,000 dollars per seat per year. The underlying models are a routing layer over OpenAI and Anthropic frontier models.

Thomson Reuters CoCounsel — the unified assistant after the Casetext acquisition

CoCounsel began as a product from a startup called Casetext, announced the day GPT-4 launched in March 2023. In June 2023, Thomson Reuters acquired Casetext for 650 million dollars. Since then CoCounsel has merged with Westlaw and Practical Law into a single large assistant platform.

Major skills:

  • Legal research memos — Westlaw data with citation verification.
  • Contract policy compliance — compare incoming contracts to your policy.
  • Contract summary — extract key clauses, risks, expiry.
  • Deposition prep — find contradictions in deposition transcripts.
  • Database exploration — natural-language queries over Westlaw, EDGAR, and others.

Pricing is around 400 dollars per user per month (roughly 4,800 dollars per user per year), with volume discounts depending on firm size. Models are a custom-tuned mix of OpenAI and Anthropic, and the key differentiator is the Westlaw KGI (Knowledge Graph Index) used to verify citations.

The Casetext legacy — the original of AI legal research

Casetext itself has been absorbed into CoCounsel and no longer exists as a standalone brand, but the legacy matters. On the day GPT-4 launched in March 2023, Casetext announced CoCounsel as the first commercial AI legal assistant. It standardized a pattern for the legal domain — retrieve first, then answer only on top of retrieved citations — that we now call retrieval-augmented generation (RAG).

That pattern was adopted by almost every later entrant: Lexis+ AI, vLex Vincent, Paxton, and others. In the domain where hallucinations are scariest, Casetext showed that index correctness beats model creativity.

Spellbook — a tool that writes contracts inside Microsoft Word

Spellbook is a Word add-in from a Canadian company called Rally Legal. The core value is simple — lawyers already draft in Word, so the AI should live inside Word. No separate web app, just a side panel inside Microsoft Word.

Headline features:

  • Clause generation and rewriting — make selected text stronger, softer, more neutral.
  • Risk flagging — automatically mark non-standard clauses.
  • Benchmark comparison — position the language against market standard.
  • Q and A — ask natural-language questions about the whole contract.
  • Spellbook Associate — a longer-context agent mode launched in 2025.

Pricing is 129 dollars per user per month (around 1,548 dollars per user per year). That is far cheaper than Harvey or CoCounsel and is aimed at midsize firms and in-house legal teams. By late 2025 Spellbook announced more than 2,500 firm customers.

Robin AI — a contract review specialist

Robin AI is a UK company. It started as a human-in-the-loop contract review service and pivoted to a fully automated platform. It is known for running Anthropic Claude as its primary model.

Features:

  • Reports — extract key clauses from uploaded contracts into a one-page report.
  • Playbook comparison — match incoming contracts against your standard positions.
  • Redlining — propose automatic edits as track changes.
  • Q and A — natural-language queries on contracts.

Marquee customers are global in-house teams (UK telecom, Japanese auto OEMs, and others), and Robin AI is especially strong in the in-house market. Pricing is negotiated by account, generally in the range of 2,000 to 5,000 dollars per seat per year.

Ironclad — how the CLM leader absorbs AI

Ironclad, founded in 2014, is a contract lifecycle management (CLM) company. It did not start as an AI company. It absorbed AI gradually into CLM. The 2024 round valued it around 3.2 billion dollars.

AI features:

  • AI Assist — contract review, summary, flagging.
  • Repository Search — natural-language search over your contract store (for example, last year's renewed NDAs with auto-renewal clauses).
  • Workflow Designer — no-code workflow design with AI calls at each step.

Competitors in CLM include LinkSquares, Lexion (acquired by Docusign in 2023), Evisort, and ContractPodAi. CLM does not sell on AI alone — it has to integrate with signatures (DocuSign, Adobe Sign) and ERP or CRM systems. That advantages established CLM companies over pure AI startups.

Luminance — unsupervised contract AI born in the UK

Luminance was founded in Cambridge in 2015. Before GPT, it ran its own transformer and unsupervised learning for contract analysis. In 2023 and 2024 it added a GPT-based chat surface called Lumi, and by 2025 it settled into a hybrid that routes between its own model and frontier LLMs.

Its strengths are multilingual coverage (more than 80 languages) and a deep index of EU and UK legal frameworks. Its weakness has been slower US penetration. A 2025 Series D pushed valuation to about 1.3 billion dollars.

Kira Systems — the classic DD tool inside Litera

Kira Systems was founded in Canada in 2011 and was the standard bearer of pre-LLM NLP. It ships with more than 4,000 pretrained "smart fields" (ownership clauses, change-of-control clauses, and so on), and it dominated M&A data room analysis. Litera Microsystems acquired Kira in 2021.

In the LLM era, Kira keeps "depth of dataset" as its differentiator against newer LLM-native entrants (Eve AI, Diligen). It still has strong share in M&A due diligence, but is losing ground in general contract review to Robin AI and Luminance.

Evisort — contract intelligence inside Workday

Evisort was founded in 2016 by Harvard-trained lawyers. Workday acquired it in 2024. It is now integrated into Workday HCM and finance modules and positioned as a platform that unifies people, money, and contracts.

Functionally it is similar to other CLM tools, but the Workday integration is decisive. Fortune 500 in-house legal teams that already use Workday tend to activate Evisort as a module rather than procure it separately.

Hebbia — search-driven intelligence

Hebbia targets not just lawyers but also private equity and hedge funds. The core is "Matrix," a spreadsheet-like cell interface where you ask the same question across hundreds or thousands of documents and answers arrive in a table. If a lawyer needs to pull "governing law clause" from 100 NDAs at once, Matrix handles it in one pass.

Series B in 2024 valued the company at 700 million dollars, with investors including OpenAI Startup Fund and Index Ventures. Usability is well reviewed, but conservative firms are slow on adoption due to citation verification and security maturity questions.

vLex Vincent — assistant on top of a global case database

vLex is a Spanish-rooted global case database company that indexes statutes and case law from more than 100 countries. The merger with UK-based Justis in 2023 pushed its scale further, and in 2024 it launched the Vincent AI assistant.

Vincent's edge is global depth of data. Unlike US- and UK-centric Westlaw or Lexis, vLex has strong indexes across Latin America, the EU, and parts of Asia. That makes it powerful for cross-border comparative research.

Westlaw Precision — Thomson Reuters' main lineup

Westlaw is one of the two pillars of US legal research (the other is LexisNexis). "Westlaw Precision with CoCounsel" layers AI summarization, citation verification, and natural-language queries on top of the existing Westlaw index.

Existing Westlaw customers can switch AI features on with an extra license, typically priced at several thousand dollars per seat per year on top of base Westlaw. AmLaw firms already spend heavily on Westlaw, so turning AI on inside Westlaw is the conservative choice compared to introducing a separate AI tool.

LexisNexis Lexis+ AI — Westlaw's eternal rival

Lexis+ AI is LexisNexis' AI assistant, beta in 2023 and generally available in 2024, now growing fast. It competes with Westlaw across nearly identical categories — research, summary, drafting — at similar price points.

Its differentiator is the global LexisNexis catalog — Halsbury's Laws in the UK, Lextenso in France, Quicklaw in Canada, and others — unified into a single multi-jurisdictional research surface. UK and EU cross-border firms tend to prefer Lexis+ AI.

Relativity aiR — the giant of e-discovery

Relativity, founded in Chicago in 2001, is the leading e-discovery platform with estimated market share in the 30 to 40 percent range as of 2026. "aiR" is its AI lineup brand. Major modules:

  • aiR for Review — automates first-pass review (responsive versus not-responsive classification).
  • aiR for Privilege — flags attorney-client privileged documents automatically.
  • aiR for Case Strategy — surfaces case strengths and weaknesses.

E-discovery is the area where AI ROI is the most obvious, because data volumes are huge (hundreds of GB to multiple TB). First-pass review labor runs 30 to 80 dollars per hour and a single case can cost hundreds of thousands of dollars. If AI removes 70 to 90 percent of that, the impact is millions of dollars per case.

Everlaw, DISCO, Logikcull, Reveal — the e-discovery field

Competition in e-discovery outside Relativity is fierce.

  • Everlaw — California-based, cloud-native, strong on AI visualization.
  • DISCO — Texas-based, IPO'd in 2021 with a bumpy ride since, doubling down on AI.
  • Logikcull — self-serve, midmarket focus, acquired by Reveal in 2023.
  • Reveal — strong predictive coding and classification after acquiring Brainspace.

Relativity is number one, but Everlaw is taking share fast in the cloud transition.

Diligen, Eve AI, Paxton AI, Patexa, Genie AI — newer entrants

New entrants are plentiful.

  • Diligen — Canadian contract analysis startup, midmarket and in-house focus.
  • Eve AI — California, focused on litigation automation.
  • Paxton AI — Washington DC, compliance and research assistant.
  • Patexa — patent analysis specialist.
  • Genie AI — UK, draft-time assistance, the UK answer to Spellbook.

These teams generally avoid head-on fights with category leaders and pick specific verticals (patents, litigation, in-house drafting) where they can establish themselves.

Pricing matrix — per-seat cost at a glance

A summary of the publicly known or widely reported per-seat pricing as of May 2026.

ToolCategoryPer-seat range
HarveyPlatform assistantOver 5,000 USD per year (enterprise)
CoCounselAssistantAround 400 USD per month
SpellbookWord drafting129 USD per month
Robin AIContract review2,000 to 5,000 USD per year
IroncladCLM1,500 to 3,000 USD per year
Westlaw PrecisionResearchBase Westlaw plus AI add-on
Lexis+ AIResearchBase Lexis plus AI add-on
Relativity aiRE-discoveryData-volume-based
LuminanceContract analysisAround 2,500 USD per seat per year
Genie AI and PaxtonDrafting and research entrants50 to 150 USD per month

Pricing varies sharply by firm size, contract length, and integration scope, so treat these as ranges.

Hallucinated citations — Mata v. Avianca and what followed

In May 2023, in the US District Court for the Southern District of New York, attorney Steven Schwartz filed a brief drafted with ChatGPT. The six cases it cited turned out to be entirely fabricated (Mata v. Avianca). Schwartz was sanctioned with a 5,000 dollar fine and severe reputational damage.

After the case, US courts began requiring verification of AI-generated material, and some judges started demanding AI-use declarations on every filing. By 2024, more than 30 additional hallucinated-citation incidents had been reported.

This risk is the decisive reason most legal AI tools moved to a RAG architecture. The standard is no longer free generation but "only allow citations pulled from a verified index."

GDPR and attorney-client privilege

Legal tech AI carries two regulatory collisions.

First, GDPR and processing location. Sending European client data to a US LLM cloud may violate the cross-border transfer rules of GDPR. Harvey and CoCounsel offer EU data residency options, but not every tool does.

Second, attorney-client privilege. When client communications are sent to an LLM provider, the provider must guarantee that the data is not used for training. The American Bar Association in 2024 issued Formal Opinion 512, stating that AI use is permitted but client notice on data handling and confidentiality is required.

Because of these two issues, conservative firms have standardized on (1) on-premise or VPC deployment, (2) zero-data-retention contracts, and (3) client-specific consent forms.

Korean legal tech ecosystem

In Korea, the Attorney-at-Law Act limits non-lawyers from operating legal services, so legal tech companies position themselves as tool providers rather than direct legal advisors.

  • LawAndCompany — runs LawTalk; after navigating tensions with the Korean Bar Association from 2024 to 2025, now has an AI assistant lineup.
  • BISCUIT — in-house LLM and contract review SaaS, targeting startups and midmarket in-house teams.
  • Caseway (global with Korean market presence) — Canadian-rooted but engages with Korean firms.
  • AI lawyer (multiple startups) — consumer-facing free consultation with clearly limited scope.
  • Firm internal tools — Kim and Chang, Lee and Ko, Bae Kim and Lee are adopting internal LLMs.

Korean-language data is smaller than English, and the case-law and statute index is less granular than in English-speaking jurisdictions. That is the main reason global tools cannot be imported as-is. A hybrid of local index plus LLM has become the local standard.

Japanese legal tech ecosystem

Japan has a bigger and more conservative market than Korea. The four largest Tokyo firms (Nishimura, Mori, Anderson Mori, TMI) have outsized influence.

  • BUSINESS LAWYERS by Bengo4.com — legal information and template platform, gradually adding AI.
  • FRAIM AI — Japanese contract review and drafting startup.
  • GVA TECH — automated contract review, targeting Japanese in-house teams.
  • Hubble — contract drafting collaboration SaaS, AI integration in progress.
  • MNTSQ — backed by Nomura and SoftBank, AI contract analysis.

The Practising Attorney Act (Bengoshi-ho) Article 72 limits non-lawyer legal practice, and Japanese in-house teams are unusually influential. That gives in-house SaaS a strong position. Japanese-language LLM performance has closed most of the gap with English since GPT-4o and 5, but tools still need post-training to handle Japanese-specific contract conventions (party labels "Kou and Otsu," "Governing law," "Stamp tax").

Open-source and academic resources

Legal NLP has an active research community.

  • LexNLP — a Python library for extracting dates, clauses, parties from legal text.
  • Lexpredict — the parent company of LexNLP, offering consulting and custom models.
  • Legal-BERT — a BERT-family model further trained on legal corpora.
  • CUAD (Contract Understanding Atticus Dataset) — 510 contracts with 41 clause labels, for training and evaluation.
  • CaseHOLD — a benchmark for predicting case holdings in US case law.
  • LegalBench — a benchmark suite of diverse legal tasks.

If you do not have budget for commercial tools, or need to build internally, these resources are good starting points. Note that they are US- and UK-centric, so Korean and Japanese data needs separate corpus work.

Adoption checklist — what firms and in-house teams should verify

Items to confirm when adopting legal tech AI.

  1. Data residency — where is EU, Korean, or Japanese client data processed?
  2. Zero data retention — is there a contract stating the provider will not train on inputs?
  3. Attorney-client privilege — are client consent and notice procedures in place?
  4. Citation verification — does the tool auto-verify citations, or must a lawyer check every time?
  5. Logs and audit — can you audit who queried what and when?
  6. On-premise or VPC — large firms should evaluate self-hosted infrastructure.
  7. Lawyer ethics (ABA Opinion 512 and equivalents) — is your usage policy documented?
  8. Training — usage training and mistake case sharing for lawyers and paralegals.
  9. ROI measurement — hours saved, dollars saved per document, post-adoption satisfaction.
  10. Phased rollout — start with low-risk tasks (summary, translation) and end with high-risk tasks (in-court advocacy).

12 to 24 month outlook — agents will run routine matters

The big trends across 2026 to 2028.

  • AI agents will run routine matters end-to-end — NDA review, the first round of standard-contract negotiation, simple compliance checks. Lawyers will supervise and approve.
  • The lawyer becomes a supervisor — less direct drafting, more review and sign-off of AI output. Hourly value holds or rises.
  • Pricing model shifts — billable hours come under pressure; value-based and fixed-fee pricing grows.
  • Junior lawyer training is both at risk and an opportunity — the work that years one through three used to do is automated, so the path to senior must be redesigned.
  • Stronger regulation — mandatory AI use declarations and retained verification records become the norm.
  • Rise of local models — data sovereignty pushes EU, Japanese, and Korean local LLM use (Mistral, LLaMA fine-tunes, Japanese and Korean foundation models), reducing US cloud dependency.
  • Long-tail consolidation — half of the 100-plus legal tech startups will be acquired or fold, with one or two leaders consolidating each category.

References

Closing — the lawyer's work does not disappear, it changes shape

By May 2026, legal tech has moved past the question of whether AI replaces lawyers. The interesting questions are two.

First, which firm rides down the AI workflow learning curve fastest, and gets the same lawyer count to process twice the matter load. A firm's competitive edge becomes matters-per-lawyer, not lawyer count.

Second, which in-house teams reduce dependency on outside counsel and bring work back in-house. AI returns parts of the standard work that outside firms monopolized to in-house. In-house influence grows; outside firms have to move up the difficulty curve.

The lawyer's work does not disappear. But its shape changes. Tasks that took a first-year three days now finish in 30 minutes, and a fifth-year's time moves to harder negotiation and strategy. Who makes that transition smoothest is what will redraw the legal market map of 2028.

Discuss on Twitter • View on GitHub