The previous version of this page ranked Casetext first. You cannot buy Casetext. Thomson Reuters retired the standalone product on April 1, 2025, and any 2026 shortlist that still opens with it is quoting a dead SKU. Here is the list that reflects what is actually purchasable, ranked for legal research specifically — finding the authority, confirming it still stands, and citing it without getting sanctioned.
The thing that separates these four in 2026 is no longer “does it hallucinate.” Every serious vendor ships citation grounding now. The split is: who owns the corpus underneath the answer, and what checks the citation after the model writes it. That question routes almost every buying decision below.
1. Thomson Reuters CoCounsel Legal — the default for US litigation research
CoCounsel Legal is what the Casetext technology became. Thomson Reuters relaunched it in August 2025 around Deep Research, an agentic research capability that takes a full research question, plans its own approach, and returns an answer grounded in Westlaw and Practical Law rather than in the open web. Deep Research Verify then re-reads the finished report and checks whether the cited authority actually supports the assertion it was attached to, flagging misattributions and mischaracterisations. It expanded to the UK on January 26, 2026. ooligo score: 8.6.
That two-step — agentic research over a licensed corpus, then a machine check on whether the citation says what the brief claims it says — is the strongest answer on this page to the failure mode that keeps ending up in sanctions orders.
What it replaces: the associate’s first-pass Westlaw session and the memo that follows it. It does not replace the associate’s judgment about which argument to make.
Where to start: take one research memo you have already written, run it through Deep Research, and compare the authority it surfaces against what your associate found. You are testing recall, not prose.
2. LexisNexis Protégé — the same shape on the other corpus
Protégé is the mirror image, and the choice between the two is mostly a choice of database. LexisNexis renamed Lexis+ AI to Lexis+ with Protégé in February 2026, then shipped Protégé Work on May 7, 2026: describe a legal goal in plain language, get a structured plan back, approve it, and it executes across 300+ skills covering contract comparison, complaint analysis, due diligence, and compliance review. Shepard’s Verify Trust Markers, also shipped May 2026, scan both AI-generated and attorney-drafted text, check every citation against Lexis sources, and flag anything they cannot verify. ooligo score: 8.0.
Protégé’s specific win is that check running over text a human wrote. CoCounsel’s verification is scoped to its own reports; Shepard’s Verify will read the brief your partner drafted by hand.
What it replaces: Lexis Advance manual research, plus the cite-check pass a paralegal used to run before filing.
Where to start: if Lexis is already your research contract, turn Protégé on for one practice area for 60 days and measure cite-check time, not research time — that is where the delta shows up first.
3. vLex Vincent AI — the multi-jurisdiction pick, and the cheapest real research seat
The prior version of this page dismissed Vincent as a “narrower realistic decision space.” That was wrong by the time it was written. Clio closed a $1 billion acquisition of vLex in November 2025, and Vincent now sits inside the practice-management platform most small and mid-size firms already run. It grounds its output in a corpus vLex puts at over 1 billion documents across 100+ countries, and it ships litigation-specific workflows — judge profiling, lawyer profiling, complaint analysis — plus Vincent Studio, a no-code builder that lets a firm encode its own research process as a repeatable workflow.
Vincent wins outright on two scopes. Cross-border work, where Westlaw and Lexis both thin out fast outside their home jurisdictions and Vincent does not. And price per research seat: vLex does not publish a rate card, but third-party reporting puts self-serve access with Vincent included around $65–$69/user/month, roughly a quarter to a tenth of what the two incumbents charge for a comparable seat. Firm plans reportedly land nearer $399/user/month. More than 40 state bars also carry the Fastcase base as a member benefit, with Vincent riding on top of whatever plan you already hold — check your bar before you buy anything.
There is no ooligo entry for Vincent yet; it is in the authoring queue, and this page will link to it when it exists.
What it replaces: a second research subscription for foreign or multi-state matters, and — for a lot of small firms — the entire premium research line item.
Where to start: take the last matter that crossed a jurisdiction and re-run the research in Vincent. If your practice never leaves one state, this is not the pick.
4. Harvey — research inside the matter, at firm scale
Harvey is the highest-scoring legal-AI platform in our catalog at 8.8, and it ranks fourth here anyway. That is not a downgrade of Harvey; it is what happens when you rank for research alone. Harvey does not own primary law. It reaches case law through a strategic alliance with LexisNexis announced in June 2025, which integrates LexisNexis primary law content and Shepard’s Citations into the Harvey platform — you ask, and a Protégé-backed service answers with validated citations inside Harvey’s interface.
That is a genuinely good outcome, and it is also the reason Harvey sits below the two vendors that own the corpus: you are buying the same authority with a partnership between you and it. What you get in exchange is research that happens where the matter lives, next to the drafting, the vault, and the workflow agents, instead of in a separate research tab.
Harvey is also the most expensive option here by a wide margin — no published rate, reported seat pricing around $1,000–$1,200/month, annual contracts from $50K upward, and 25–50 seat minimums that rule it out below roughly 50 attorneys regardless of budget.
What it replaces: outside-counsel hours on first-pass research and drafting, at firms where research is one of several AI use cases rather than the only one.
AI add-on alone ~$125–$275/mo; ~$99 per AI Ask transactional
TR CoCounsel Legal
~$225/mo (Core) to ~$639/mo (Westlaw Advantage, solo)
Online tiers stop at 10 attorneys
Harvey
~$1,000–$1,200/mo
25–50 seat minimum; $50K+ annual floor
The spread between the cheapest and most expensive seat on this page is roughly 15×. None of the vendors publish a clean rate card, so treat every figure above as a reported band to negotiate against, not a price. The one durable rule: if you already pay for Westlaw or Lexis, the AI layer is an add-on line item on top of that contract, not a replacement for it — budget both.
What’s not on this list and why
Casetext — retired April 1, 2025. Kept here only so that former subscribers landing on old reviews know where the product went: into CoCounsel Legal, at roughly two to three times the old standalone rate.
Westlaw or Lexis without the AI layer — still the corpus underneath two options above, and still a reasonable purchase. They are not AI research tools, and buying one does not get you Deep Research or Protégé.
Bloomberg Law AI — real product, but its research pull is concentrated in transactional and regulatory work where Bloomberg’s own analytics are the draw. If Bloomberg Law is already your contract, evaluate its AI in place; nobody switches to it for the AI.
Spellbook and Legora — strong tools that belong on a contracts shortlist, not a research one. Neither ships a citator.
Claude and ChatGPT — never alone for anything that gets filed. They are the correct tool for reasoning on top of authority you have already verified: pressure-testing an argument, restructuring a memo, summarizing a transcript. Use them after the citation check, not instead of it.
Minimum viable choice
If you cannot justify a full evaluation, pick by the contract you already have:
Already on Westlaw → CoCounsel Legal. Continuity of corpus is worth more than a feature delta.
Already on Lexis → Protégé. Same logic, other side.
Neither, or under ten attorneys, or work crosses jurisdictions → vLex Vincent AI. This is the default if you are choosing in a vacuum. It is the cheapest credible research seat on the page and the only one you can start using this week without a procurement cycle.
50+ attorneys with AI use spread across practice areas → Harvey, and accept that you are buying a platform, not a research tool.
If none of the four fit — usually because the budget is zero — the honest status quo is your bar association’s Fastcase benefit plus manual cite-checking. That is not a good workflow. It is a real one, and it beats an unverified LLM citation by every measure that matters.
When to revisit
Re-check this page if any of three things happen: Clio finishes folding Vincent into the core Clio subscription (which would change the price comparison above materially), Thomson Reuters or LexisNexis publishes an actual rate card (neither has), or the Harvey–LexisNexis alliance changes shape — Harvey’s research position on this page rests entirely on it.
The previous version of this page ranked Casetext first. You cannot buy Casetext. Thomson Reuters retired the standalone product on April 1, 2025, and any 2026 shortlist that still opens with it is quoting a dead SKU. Here is the list that reflects what is actually purchasable, ranked for legal research specifically — finding the authority, confirming it still stands, and citing it without getting sanctioned.
The thing that separates these four in 2026 is no longer “does it hallucinate.” Every serious vendor ships citation grounding now. The split is: who owns the corpus underneath the answer, and what checks the citation after the model writes it. That question routes almost every buying decision below.
1. Thomson Reuters CoCounsel Legal — the default for US litigation research
CoCounsel Legal is what the Casetext technology became. Thomson Reuters relaunched it in August 2025 around Deep Research, an agentic research capability that takes a full research question, plans its own approach, and returns an answer grounded in Westlaw and Practical Law rather than in the open web. Deep Research Verify then re-reads the finished report and checks whether the cited authority actually supports the assertion it was attached to, flagging misattributions and mischaracterisations. It expanded to the UK on January 26, 2026. ooligo score: 8.6.
That two-step — agentic research over a licensed corpus, then a machine check on whether the citation says what the brief claims it says — is the strongest answer on this page to the failure mode that keeps ending up in sanctions orders.
What it replaces: the associate’s first-pass Westlaw session and the memo that follows it. It does not replace the associate’s judgment about which argument to make.
Where to start: take one research memo you have already written, run it through Deep Research, and compare the authority it surfaces against what your associate found. You are testing recall, not prose.
Full Thomson Reuters CoCounsel review →
2. LexisNexis Protégé — the same shape on the other corpus
Protégé is the mirror image, and the choice between the two is mostly a choice of database. LexisNexis renamed Lexis+ AI to Lexis+ with Protégé in February 2026, then shipped Protégé Work on May 7, 2026: describe a legal goal in plain language, get a structured plan back, approve it, and it executes across 300+ skills covering contract comparison, complaint analysis, due diligence, and compliance review. Shepard’s Verify Trust Markers, also shipped May 2026, scan both AI-generated and attorney-drafted text, check every citation against Lexis sources, and flag anything they cannot verify. ooligo score: 8.0.
Protégé’s specific win is that check running over text a human wrote. CoCounsel’s verification is scoped to its own reports; Shepard’s Verify will read the brief your partner drafted by hand.
What it replaces: Lexis Advance manual research, plus the cite-check pass a paralegal used to run before filing.
Where to start: if Lexis is already your research contract, turn Protégé on for one practice area for 60 days and measure cite-check time, not research time — that is where the delta shows up first.
Full LexisNexis Protégé review →
3. vLex Vincent AI — the multi-jurisdiction pick, and the cheapest real research seat
The prior version of this page dismissed Vincent as a “narrower realistic decision space.” That was wrong by the time it was written. Clio closed a $1 billion acquisition of vLex in November 2025, and Vincent now sits inside the practice-management platform most small and mid-size firms already run. It grounds its output in a corpus vLex puts at over 1 billion documents across 100+ countries, and it ships litigation-specific workflows — judge profiling, lawyer profiling, complaint analysis — plus Vincent Studio, a no-code builder that lets a firm encode its own research process as a repeatable workflow.
Vincent wins outright on two scopes. Cross-border work, where Westlaw and Lexis both thin out fast outside their home jurisdictions and Vincent does not. And price per research seat: vLex does not publish a rate card, but third-party reporting puts self-serve access with Vincent included around $65–$69/user/month, roughly a quarter to a tenth of what the two incumbents charge for a comparable seat. Firm plans reportedly land nearer $399/user/month. More than 40 state bars also carry the Fastcase base as a member benefit, with Vincent riding on top of whatever plan you already hold — check your bar before you buy anything.
There is no ooligo entry for Vincent yet; it is in the authoring queue, and this page will link to it when it exists.
What it replaces: a second research subscription for foreign or multi-state matters, and — for a lot of small firms — the entire premium research line item.
Where to start: take the last matter that crossed a jurisdiction and re-run the research in Vincent. If your practice never leaves one state, this is not the pick.
4. Harvey — research inside the matter, at firm scale
Harvey is the highest-scoring legal-AI platform in our catalog at 8.8, and it ranks fourth here anyway. That is not a downgrade of Harvey; it is what happens when you rank for research alone. Harvey does not own primary law. It reaches case law through a strategic alliance with LexisNexis announced in June 2025, which integrates LexisNexis primary law content and Shepard’s Citations into the Harvey platform — you ask, and a Protégé-backed service answers with validated citations inside Harvey’s interface.
That is a genuinely good outcome, and it is also the reason Harvey sits below the two vendors that own the corpus: you are buying the same authority with a partnership between you and it. What you get in exchange is research that happens where the matter lives, next to the drafting, the vault, and the workflow agents, instead of in a separate research tab.
Harvey is also the most expensive option here by a wide margin — no published rate, reported seat pricing around $1,000–$1,200/month, annual contracts from $50K upward, and 25–50 seat minimums that rule it out below roughly 50 attorneys regardless of budget.
What it replaces: outside-counsel hours on first-pass research and drafting, at firms where research is one of several AI use cases rather than the only one.
Full Harvey review →
Pricing reality
The spread between the cheapest and most expensive seat on this page is roughly 15×. None of the vendors publish a clean rate card, so treat every figure above as a reported band to negotiate against, not a price. The one durable rule: if you already pay for Westlaw or Lexis, the AI layer is an add-on line item on top of that contract, not a replacement for it — budget both.
What’s not on this list and why
Minimum viable choice
If you cannot justify a full evaluation, pick by the contract you already have:
If none of the four fit — usually because the budget is zero — the honest status quo is your bar association’s Fastcase benefit plus manual cite-checking. That is not a good workflow. It is a real one, and it beats an unverified LLM citation by every measure that matters.
When to revisit
Re-check this page if any of three things happen: Clio finishes folding Vincent into the core Clio subscription (which would change the price comparison above materially), Thomson Reuters or LexisNexis publishes an actual rate card (neither has), or the Harvey–LexisNexis alliance changes shape — Harvey’s research position on this page rests entirely on it.