This article is part of our AI Agent Guides. See how models compare this week on the AI Agent Leaderboard, a ranking of rankings refreshed every week.
0 entries
The best AI agent for a law firm depends on your size and practice. Look at Harvey, Legora, or CoCounsel Legal for larger firms, and Clio or Paxton for small firms. Spellbook suits contract work, EvenUp or Eve suit plaintiff firms, and Smith.ai answers intake calls. Every one of them still needs a lawyer checking the output.
This article is general information and not legal advice. A qualified attorney must review any AI output before it reaches a client or a court. State bars may add their own rules, so check yours.
Quick comparison
| Tool | Best for | Starting price (billing) | Key limit |
|---|---|---|---|
| Harvey | Large firms and in-house teams | Contact sales | No public pricing |
| CoCounsel Legal | Firms relying on Westlaw and Practical Law | Contact sales for law firm plans | Value depends on the Thomson Reuters content you license |
| Lexis+ with Protégé | Firms relying on Lexis content | Contact sales | Pricing depends on firm size and subscription |
| Legora | Larger firms running multi-step work | Contact sales | No public pricing |
| Clio | Solo and small firms | Starter $49 per user | AI is on plans from Core up, and those need a quote |
| Paxton | Solo attorneys wanting research and drafting | $499 per user per month, or $2,999 per year | One published individual plan price |
| Spellbook | Transactional lawyers in Word | Contact sales after 7-day trial | Focused on contracts, not litigation |
| EvenUp and Eve | Personal injury and plaintiff firms | Contact sales | Narrow practice focus |
| Smith.ai AI Receptionist | Answering and qualifying calls | $0 for 25 calls a month | $3 per call after that on the free plan |
Prices checked on vendor pages on 2 November 2026.
How we chose
- We read vendor documentation and pricing pages. We did not run our own tests.
- Tools had to do multi-step legal or firm work, not just chat.
- We looked for stated data terms, such as whether your inputs train models.
- We covered different firm sizes and practice areas on purpose.
- We read ABA Formal Opinion 512 to frame the checks you need.
- The numbering is not a ranking. For underlying model positions, see the leaderboard.
What ABA Opinion 512 means for your purchase
The ABA issued its first formal opinion on generative AI on 29 July 2024. It says lawyers must consider duties of competence, confidentiality, client communication, supervision, candor, and reasonable fees. Those duties shape how you should evaluate a tool.
The opinion says relying on AI output without appropriate independent verification or review could violate the competence duty in Model Rule 1.1. It adds that the amount of review depends on the tool and the task. So build review time into your budget from the start.
It is a national model opinion, so check your own state's guidance too. Your firm's partners, not a vendor, are the ones who decide what is acceptable.
1. Harvey: the large-firm platform
Best for: Large law firms, in-house teams, and professional service firms.
Price: Contact sales. Harvey's site shows no public price and offers ROI calculators instead.
Harvey Agents split large tasks, work in parallel, and return review-ready output. The platform also lists contract intelligence, regulatory horizon scanning, intake, and a connector library that includes custom MCP. MCP is an open standard for linking AI tools to other software.
On data, Harvey's security page says its agreement guarantees inputs, outputs, and uploads are not used to train underlying models. It also lists retention controls and in-region storage in the EU or Switzerland, the US, or Australia. Those are the terms a firm's risk team asks about first.
Where it falls short: No public pricing, and a smaller firm is unlikely to need its full range. Its breadth also means change management is a real project, not a quick add.
Our take: A leading candidate for large firms with an innovation team. If you are a small firm, you are probably paying for capabilities you will not use.
2. CoCounsel Legal: for Westlaw firms
Best for: Firms that already rely on Westlaw and Practical Law.
Price: Contact sales. Thomson Reuters' plans page directs law firms to a sales representative.
CoCounsel Legal handles research, analysis, and drafting. It uses agentic AI grounded in Westlaw content, Practical Law guidance, and your firm's documents. Thomson Reuters says drafts come with linked citations for verification.
The advantage is the content. A tool answering from licensed primary law and editorial guidance gives you a clearer path to checking sources. Linked citations make the Rule 1.1 review faster, but they do not replace reading the cases.
Where it falls short: Pricing is not public, and the tool is most valuable if you are already invested in Thomson Reuters content. Grounded AI can still make mistakes, so citation checks remain your job.
Our take: The obvious first trial if Westlaw is your research home. Compare it to your current workflow on a matter you already finished, so you can spot errors.
3. Lexis+ with Protégé: for Lexis firms
Best for: Firms that already use Lexis for research.
Price: Contact sales. LexisNexis says pricing is customized by organization size, practice needs, and subscription requirements.
Protégé combines extractive, generative, and agentic AI for research, drafting, and document analysis. LexisNexis also lists custom agent skills and describes work with Microsoft to bring its legal AI into Microsoft 365 Copilot. Lexis+ with Protégé pairs the assistant with the Lexis research platform.
The choice between this and CoCounsel often comes down to which content your lawyers trust and already pay for. Switching research platforms for an AI feature is rarely worthwhile on its own. Run both on the same sample tasks if you are genuinely undecided.
Where it falls short: No public pricing, so comparing it with CoCounsel takes two sales cycles. Like every tool here, it does not remove the need to verify authority.
Our take: A natural choice for Lexis firms. Ask how agent skills are reviewed and who in your firm can publish one.
4. Legora: a large-firm alternative to Harvey
Best for: Larger firms and legal teams that want an agent for end-to-end work.
Price: Contact sales. Legora's site shows no public price.
Legora describes an agent that plans, executes, reviews, and delivers complex legal work, leaving judgment calls to lawyers. It also offers Monitors, which scan global regulation for changes. Its broader system connects legal information, communication, and execution.
Legora and Harvey compete for many of the same buyers. Include both if you are running a formal evaluation, since a competing proposal also helps your negotiation. Ask each to work on the same anonymized matter.
Where it falls short: The home page gives little detail on data terms and limits, so ask for the documents. No public pricing.
Our take: Worth including in a large-firm shortlist. Ask for references from firms of your size and in your jurisdictions.
5. Clio: AI inside practice management
Best for: Solo and small firms that want AI in the system they already use to run the firm.
Price: Clio's pricing page shows Starter from $49 per user. AI is included on plans from Core up, and other tiers show "Contact sales for pricing."
Clio's product range now spans Clio Manage for running the practice, Clio Grow for intake, and Clio Work for legal research and drafting. Enterprise firms get Vincent, its legal AI for larger firms. The Manage AI walkthrough shows AI working inside matters and billing.
The appeal is context. AI that already sees your matters, deadlines, and billing needs less copying and pasting. For a small firm with no IT staff, one system can also mean one fewer vendor review.
Where it falls short: The starter price does not include AI, so the real cost requires a quote. Clio's AI spans several product names, so make sure you know what your quote covers.
Our take: The sensible starting point for small firms already on Clio. Get a written answer about which AI features are in which plan.
6. Paxton: legal research with a public price
Best for: Solo attorneys and small firms who want research and drafting without a sales call.
Price: Paxton's pricing page lists $499 per user per month or $2,999 per user per year, with a 7-day free trial.
Paxton lists research across 30,000+ legal sources, court-formatted drafting, and facts and timelines across a matter. It connects to Box, Dropbox, OneDrive, and SharePoint. It says your data never trains AI models and lists SOC 2, ISO 27001, and HIPAA on the same page.
The annual price is far below twelve monthly payments. Use the trial to decide, then pay annually if the tool earns a place. A month-to-month rate this high is only sensible for a short project.
Where it falls short: The price is high for one user on monthly billing. Compare its source coverage with your current research platform for your jurisdiction before relying on it.
Our take: Refreshing public pricing for a small firm that wants to test agentic research. Run it against a research memo you have already written.
7. Spellbook: contract drafting in Microsoft Word
Best for: Transactional lawyers and in-house teams who live in Word.
Price: Contact sales. Spellbook's pricing page describes custom pricing and a 7-day free trial.
Spellbook's Word add-in covers review, drafting, questions, benchmarks, and playbooks. Its Associate agent works through multi-document matters with your oversight. Working inside Word means lawyers can adopt it without changing how they edit documents.
Contract work suits AI better than many legal tasks, because playbooks give clear rules to check against. Write your firm's positions down first. The more specific your playbook, the more useful the suggested redlines will be.
Where it falls short: It is focused on contract work and is not a litigation research tool. The custom pricing also means you will need a sales conversation.
Our take: The practical pick for deal lawyers. Test on a contract type you see weekly, with your own fallback positions loaded.
8. EvenUp and Eve: built for plaintiff firms
Best for: Personal injury and other plaintiff firms.
Price: Contact sales for both. EvenUp and Eve use demo or scheduled-call sales.
EvenUp focuses on personal injury, with tools across intake, treatment, demands, negotiation, discovery, and trial. It also sells Communication Agents to reduce administrative work with clients. Eve covers labor and employment, personal injury, workers' compensation, disability, and medical malpractice, and includes an AI voice agent for intake.
Specialist tools know the documents these firms handle most: medical records, demand letters, and case evaluations. That usually gets you useful results faster than configuring a general tool. Case data here is highly sensitive, including medical records, so read the data terms closely.
Where it falls short: Neither helps much outside plaintiff work. Vendor case studies on time saved describe particular firms, so do not budget around them.
Our take: If plaintiff work is all you do, start here rather than with a general legal platform. Compare both on the same real file.
9. Smith.ai AI Receptionist: never miss an intake call
Best for: Small firms that lose prospective clients to voicemail.
Price: Smith.ai's plans include free with 25 calls a month and $3 per call after that. Pro is $150 a month, at $2 per call.
The agent answers 24/7, qualifies leads, routes calls, and schedules. Smith.ai lists a native Clio integration. It also offers hybrid and human-first plans if you want people on sensitive calls.
For firms, intake is the job with the clearest return. It is also lower risk than legal drafting, because the agent collects information rather than giving advice. Instruct it never to offer an opinion on a caller's case.
Where it falls short: An AI intake call can raise questions about forming a client relationship, conflicts, and recording consent. Have a lawyer write the script and review the transcripts for the first few weeks.
Our take: A quick win for small firms. See our law firm marketing services if the bigger problem is not enough calls.
How to choose
Sort by size first. Large firms should compare Harvey, Legora, and whichever research platform they already use. Small firms should start with the tool attached to what they use daily, then add one specialist.
Before any trial, ask each vendor these questions:
- Are inputs or outputs used to train models? Is that in the contract?
- Where is data stored, and how long is it kept?
- Can we set retention rules by matter or by client?
- How do citations link to a source we can check?
- What audit trail shows who ran what and when?
- What happens to our data if we leave?
Then deal with client communication and fees. Opinion 512 covers both, so decide when clients are told about AI use and how you will bill for time saved. Have your ethics counsel sign off.
Test on closed matters where you know the right answer. Our agent evaluation scorecard helps you record the results. Set approval rules too, using our human approval decision table.
Model benchmarks provide some context on the engines behind these products. The knowledge work board shows current positions, and the methodology page explains them. They do not measure legal accuracy in your jurisdiction.
Frequently asked questions
Do we have to tell clients we use AI?
Opinion 512 treats client communication as one of the duties to consider, but the answer depends on the tool and the matter. Inputting confidential information raises separate informed consent questions under Rule 1.6. Ask ethics counsel and check your state's guidance.
Can AI agents be trusted with case citations?
Not without checking. Linked citations make review faster, but Opinion 512 says lawyers need an appropriate degree of independent review of output. Read the authority yourself before filing anything.
What is the cheapest legal AI to try?
Smith.ai's free tier covers phone intake. Paxton and Spellbook offer 7-day trials. Clio's Starter plan is from $49 per user, but AI starts on higher plans that need a quote.
Is a general tool like ChatGPT enough for a law firm?
It can help with general drafting and summaries, under the right data terms. It does not come with licensed legal content or the firm controls found in legal products. Check confidentiality terms before entering client information.
How much review time should we plan for?
Opinion 512 says it depends on the tool and task, and it does not give a number. Measure it during your trial and count it as a cost. Review time often decides whether a tool saves money at all.
