Paper Instruments

AI tools for lawyers: which one fits which job

Lawyers weighing AI face four shapes of tool: research platforms grounded in Westlaw and Lexis, contract drafting in Word, AI inside practice management, and general agents. This guide prices each where a price is public and says what work fits it, with Ultramarine Work first because it is ours.

Checked on

At a glance

ProductBest forAvailable onStarts atFree access
HarveyLitigation, transactional work and diligence, large firmsWeb platform, Word and OutlookBy quoteNone
CoCounsel LegalResearch and drafting on citable Westlaw authorityWeb platform, Word, iManage, NetDocuments$1,694.90 a month on the in-house pageFree trial on request
Lexis+ with ProtégéResearch, drafting and Shepard’s validation on LexisWeb platformBy quoteFree trial, store for 1-3 attorney firms
Ultramarine WorkDrafting, research and file work in one practice modeBrowser, plus a desktop app for Mac and Windows$5 a month7 days free on Starter and Plus
Clio Work and Manage AIPractice management plus AI research and draftingWeb apps$49 a user a month7-day trial, no card
SpellbookContract drafting, redlining and playbooks in WordMicrosoft Word add-in, web workflowsBy quote7-day free trial
Claude CoworkGeneral agent work, extended with a legal pluginDesktop, web, mobile, Chrome$20 a monthPaid plans only
ChatGPT WorkLong multi-step tasks that end in finished filesDesktop, web, mobile$20 a monthLimited Work, desktop only, on free and Go

The tools to weigh

  1. 1. Harvey

    Harvey

    Good for
    Harvey is a platform for large firms and law departments: Agents, Vault, Knowledge and Spaces for matter workspaces, with apps for Contract Intelligence, Intake and Facts & Chronologies. Harvey II, released August 18, 2026, added Memory and Templates, and Harvey says it serves more than 3,500 legal organizations.
    How it works
    An enterprise web platform with Word and Outlook integrations, sold firm-wide with governance controls.
    Runs on
    Multi-model: OpenAI, Anthropic and Google since 2025. Harvey Tenet, its first model, is an August 2026 research preview, post-trained on Moonshot’s Kimi K3.
    Price
    Not public, and there is no trial or self-serve path. The site routes buyers to a demo.
    Choose it when
    Your firm is large or mid-size and is rolling out one platform across practice groups, with matters on large document sets.

    Sources: Harvey, product site, Harvey blog, Introducing Harvey II, Harvey blog, Harvey Tenet post-training

  2. 2. CoCounsel Legal

    Thomson Reuters

    Good for
    CoCounsel Legal is agentic AI grounded in Westlaw and Practical Law content. Deep Research reasons through a matter, Brief Builder assembles an argument from citable authority, and Tabular Analysis runs questions across up to 10,000 documents. It reaches Word, iManage and NetDocuments, with lighter Essentials bundles.
    How it works
    A web platform with Office integrations, priced around an existing Westlaw relationship.
    Runs on
    Not disclosed.
    Price
    Configurator pricing on 1-, 2- or 3-year terms. The in-house page lists CoCounsel Legal starting at $1,694.90 a month, with a free trial on request.
    Choose it when
    The deliverable is a memo, brief or motion that must stand on citable law, and the firm already pays for Westlaw.

    Sources: Thomson Reuters, CoCounsel Legal, Thomson Reuters, CoCounsel plans pricing

  3. 3. Lexis+ with Protégé

    LexisNexis

    Good for
    Lexis+ with Protégé, the February 2026 rename of Lexis+ AI, pairs the Lexis corpus with the Protégé assistant: research, drafting and Shepard’s citation checking. A May 2026 expansion added agentic skills, Protégé Agentic Drafting, Workrooms shared with clients and co-counsel, and Shepard’s Verify Trust Markers.
    How it works
    A web research platform, the AI layer inside an existing Lexis subscription.
    Runs on
    Third-party models from OpenAI, Anthropic and Google, per the US page. The UK page lists OpenAI and Anthropic only.
    Price
    Not public, priced by quote. LexisNexis points 1-3 attorney firms to its online store, offers a free trial and runs a small-law page.
    Choose it when
    Your firm already pays for Lexis+ and the work is research a court will read, with Shepard’s behind every answer.

    Sources: LexisNexis, Lexis+ with Protégé, LawNext, Lexis+ with Protégé expansion (May 2026)

  4. 4. Ultramarine Work

    Paper Instruments

    Ultramarine Work fits the solo or small-firm practitioner whose week crosses practice areas. Its Solo & Small-firm General Practice mode carries 23 skills, covering conflict checks, IOLTA reconciliations, demand letters, lease review and bankruptcies. Every file is a draft the lawyer reviews before it leaves the office. Research against citable authority belongs on Westlaw or Lexis, contract drafting in Word on Spellbook, firm-wide rollouts on an enterprise platform.

    A mode is a written configuration for one workstream: who the work is for, and the domain skills with their methods and review standards. Solo & Small-firm General Practice ships 23 skills, the workstream keeps the mode, and the agent draws the method each step needs. Underneath runs Ultramarine 1 Flash, our model, post-trained on GLM-5.3-Flash for knowledge work. The skills tell it to cite the file behind each figure and note what it could not verify. Tasks run Standard, Assisted with a frontier model advising, or Advanced, where a frontier model does the whole task.

  5. 5. Clio Work and Manage AI

    Clio

    Good for
    Clio’s AI comes in two layers. Manage AI, the successor to Clio Duo, drafts bills from the firm’s billing parameters and turns documents into calendar events and tasks it confirms against the source. Clio Work is a separate workspace for research, analysis and drafting on the vLex corpus, standalone since April 2026.
    How it works
    Two web apps: an assistant inside Clio Manage, plus a standalone research and drafting workspace built on vLex.
    Runs on
    Not disclosed.
    Price
    The Clio platform starts at $49 a user a month, Clio Work launched at $199 in October 2025 with no current price published, and a 7-day trial needs no card.
    Choose it when
    Your matters, billing, trust accounting and calendar already run in Clio, and you want AI inside that system.

    Sources: Clio, pricing, LawNext, Clio Work standalone (April 2026)

  6. 6. Spellbook

    Spellbook

    Good for
    Spellbook is a contract AI suite that runs inside Microsoft Word. Review redlines and flags risk, Draft builds agreements, Playbooks encode the firm’s standards, and Benchmarks compare language against market norms. Associate works across multiple documents, and ACM is in early access.
    How it works
    A Microsoft Word add-in plus web workflows.
    Runs on
    No models named on the pricing page. Spellbook says it routes to GPT-5, Claude and other models, and announced GPT-5 availability in August 2025.
    Price
    Not public. Pricing is quoted by the number of team members on the license, with a 7-day free trial.
    Choose it when
    Your week is contract redlines in Word, and you want drafting, playbook review and clause benchmarks inside the document.

    Sources: Spellbook, pricing, Spellbook, product site

  7. 7. Claude Cowork

    Anthropic

    Good for
    Claude Cowork, now folded into Claude, takes multi-step tasks across files, connectors and a browser and returns finished documents and decks. It is general by default and extended with plugins, including Anthropic’s legal role plugin, and a plugin can bundle skills, connectors and subagents.
    How it works
    A general agent on Anthropic’s own models, one assistant with optional role plugins, on desktop, web, mobile and Chrome.
    Runs on
    Anthropic’s own Claude models, including Opus, Sonnet and Fable. The Cowork page names no default model.
    Price
    Included on paid plans from Pro at $20 a month ($17 billed annually). Max is $100 and Ultra $200. Free, Pro and Max use 5-hour and weekly windows.
    Choose it when
    One assistant should cover legal and non-legal work, and the firm will build its own legal configuration in plugin files it can edit and share.

    Sources: Anthropic, Claude pricing, Anthropic Help Center, Claude Cowork, Anthropic, Cowork plugins

  8. 8. ChatGPT Work

    OpenAI

    Good for
    ChatGPT Work, the July 2026 successor to ChatGPT’s agent mode, carries long tasks to a finished file, documents, spreadsheets and decks, with cloud sessions that run in the background. For a lawyer it is the generalist option for summarizing document sets and drafting where the firm supplies the authority.
    How it works
    A general agent inside ChatGPT, on desktop, web and mobile.
    Runs on
    OpenAI’s own models, GPT-6.1 Sol and the GPT-6 line, as of November 2026.
    Price
    Plus $20 a month, Pro from $100, Business $25 a seat monthly or $20 billed annually with a 2-seat minimum. Free and Go carry limited Work, desktop only.
    Choose it when
    Your firm already works in ChatGPT and wants one general tool for drafting, document analysis and spreadsheets at a self-serve price.

    Sources: ChatGPT, plans and pricing, OpenAI Help Center, ChatGPT Work and Codex, MacRumors, ChatGPT Work launch (July 2026)

Pricing and how to try it

Plans start at $5 a month, and Starter and Plus begin with 7 days free. On our internal PI Knowledge Work benchmark, Ultramarine 1 Flash costs about $0.13 a task, about a tenth of Claude Opus 5.5 or GPT-6 Sol at comparable scores.

Start free trial See every plan

Questions

Is it safe to give an AI tool confidential client documents?

Read each vendor’s terms before uploading anything that identifies a client. Spellbook publishes zero data retention agreements and SOC 2 Type II compliance, and Thomson Reuters and LexisNexis work inside enterprise contracts. Check whether a general agent is on consumer or business terms. Ultramarine Work’s model runs on Paper Instruments’ servers, so the content it works on is sent there to be processed. Paper Instruments is SOC 2 Type II, and our security and compliance details are on our compliance page.

Can AI-generated citations be trusted in a memo or brief?

No, not until a lawyer has checked each one. The lawyer who signs it answers for every citation, whatever tool produced it. Ultramarine Work’s skills tell it to cite its sources and note what it could not verify, and the lawyer still confirms every authority. Westlaw and Lexis platforms start from citable material, not from a model’s memory.

Should a solo lawyer buy a specialist legal tool or a general assistant?

Start from the work that dominates the week. Research against citable authority belongs on CoCounsel Legal or Lexis+ with Protégé, contract drafting in Word on Spellbook, and matters, billing and calendaring in Clio. A general assistant carries the spread of a general practice, the collection letter beside the estate plan, but nothing it produces is citable until a lawyer checks it. Many firms run one of each, and mixing them is normal.

Can these tools draft court filings I can sign?

They draft, and you verify, review and sign. Whatever tool produced a filing, the attorney who signs it answers to the court for the citations and record references in it. Ultramarine Work drafts papers, and its skills have it check them against the local rules and standing orders you give it, but it does not sign, file or submit anything, and it does not touch court e-filing.

What should legal AI cost a small firm?

Published entry prices run from $5 a month for Ultramarine Work to $1,694.90 a month on CoCounsel Legal’s in-house page, and Harvey and Spellbook quote. Claude Cowork and ChatGPT Work start at $20, and Clio’s platform at $49 a user, with Clio Work’s October 2025 launch price of $199 and no current one published. CoCounsel Legal, Clio, Spellbook and Lexis+ offer trials, as do Ultramarine Work’s Starter and Plus plans, while Harvey offers none.

How do I keep control of what the AI does?

The person decides, and the tool works on your files rather than on your systems. Ultramarine Work drafts, builds and checks, changes files only in the project’s folder, sends, files and submits nothing, and does not make the professional decision. You review every file it touched before relying on it. Agents that reach into connected apps take more oversight, and the enterprise surfaces of Claude, ChatGPT and Microsoft 365 Copilot add approvals and audit controls.

Sources

Checked on