Lawyers test Claude on Pro, run out after a couple of documents, and jump to Max without knowing what they actually need. How to choose between Pro, Max, Team, the API and Enterprise for a law firm, by usage, cost and what each one does with client data. Checked October 2026.
[ KEY TAKEAWAYS ]
One attorney finding out whether Claude helps: Pro. One person reviewing documents all day: Max. Two or more people at the firm: Team, which starts at two seats, keeps your content out of model training by default and gives an admin control over connectors. A workflow that runs without a person, or privileged material at volume: the API under a zero data retention agreement. Audit logs, retention controls you set, a BAA and provisioning: Enterprise, from 20 seats. The plan decides the data path, so settle it before anyone connects Claude to the practice management system.
seats is where Team starts. Standard seats $20 a month on annual billing, Premium $100 with five times the usage, mixed as you like, up to 150.
seats is the Enterprise minimum self-serve, 50 through sales. The seat fee buys access; usage bills on top at API rates.
is how long a deleted conversation can stay in back-end storage, and the default deletion window for API inputs and outputs. Zero data retention is a separate agreement.
law firms on our last ten discovery calls already paid for Claude before the call. Two on Team, one on Enterprise, the rest on individual plans and asking how to get the firm on it.
Almost every firm we talk to has the same story. Someone starts testing Claude on Pro, is impressed, then hits a wall fast: a couple of long records in one session and the limit is reached. They upgrade to Max to keep working, and only later ask the question they should have started with: which plan does the firm need, and what does each one do with privileged client data?
This is a guide to that decision. Whether Claude is good for legal work isn't in question anymore; all ten firms on our last ten calls were paying for it before we spoke. The question is matching the plan to how the firm uses it, because the plans differ on the three things a law firm cares about: how much you can run, what it costs per person, and what happens to the data.
Pro is the cheapest paid plan and the right way to find out whether Claude helps your practice. $20 a month, or $17 on an annual plan billed up front. The catch is the usage allowance, the lowest of the paid tiers, measured in rolling five-hour windows with a weekly cap on top. Long legal documents are token-heavy, so two medical records or one deposition in a single session can hit the limit. That's the plan doing its job as an entry tier. Anthropic now sells usage credits for the moment you run out, which tells you where Pro ends: the day you buy credits twice in a month, Max is cheaper.
Max comes in two levels, $100 and $200 a month, with five or twenty times Pro's usage, monthly billing only. For a single attorney or paralegal doing real document review in the Claude app this is usually the right plan, and it adds higher output limits and priority access at busy times. Model training is opt-out on Max, the same setting as Pro, so turn it off on day one.
One distinction worth getting right: turning off training is a different thing from retention. Opting out means your content is never used to improve the model. It says nothing about how long a conversation is stored, and on the individual plans you get Anthropic's default policy with no controls over it. For internal, non-privileged work that is fine. For privileged client content, read the next two sections before you decide Max is enough.
Team is the plan for two or more people, up to 150 seats. Standard seats cost $20 a month on annual billing or $25 monthly, Premium seats $100 or $125 with five times the usage, and you mix them: Premium for the two people who draft all day, Standard for everyone else. The terms change at this tier. Anthropic does no model training on your content by default, so there is no setting for each attorney to remember. Single sign-on is included. An admin decides which remote and local connectors the firm can use, which is the control that matters once Claude starts reading matters. Claude Code and Cowork are in, and the plan pools up to $500 a month in API credits for the first automated job.
What Team doesn't have: audit logs, custom retention controls, and zero data retention. A conversation stays in the product until someone deletes it, and a deleted one leaves back-end storage within 30 days. ZDR is not a Team feature, however nicely you ask. Two more facts for the connector decision. The official MyCase connector that shipped in September installs on Team and Enterprise only, so a MyCase firm on individual plans can't use it. Our open-source Clio connector runs on any plan, Pro and Max included, because it runs on your machine with your own Clio login, and the audit log Team lacks is the one the connector writes.
Access and retention Whatever you send Claude reaches Anthropic's model for the time it takes to answer, on every plan. That is the access event, and no plan changes it. Training exclusion and retention settings govern what happens afterwards. Keep the two apart in the firm's AI policy, and treat any claim that nothing reaches a hosted model as wrong.
"I pay for the protection of Claude, I pay for Clio, as long as those are the only third parties." A litigator on a September call, on the rule his firm applies to every AI tool. The plan is what decides, in writing, what the protection of Claude amounts to.
The API is pay-per-token access that software connects to, with no seat to log into. For a firm it matters for four reasons. The largest context window, so you can hand it long documents without the app-tier limits. Zero data retention by agreement at the organization level, with no training on API content by default; for privileged data this is the clean answer. Automation: a repeatable workflow, summarize every new document, draft from a matter, post the time entries, runs on the API instead of through a person pasting into a chat. Cowork's scheduled tasks are on every paid plan, but only with remote connectors. And it connects to your practice management system: an open-source connector lets it read and write your Clio data through a controlled interface instead of re-uploading files by hand. If you are weighing this against Clio's own AI, Clio and Claude covers what each one sees.
Two things to manage. Per-token billing runs up if you feed huge documents without prompt caching and spend limits; that is an engineering detail and the reason to set it up properly, not a reason to avoid the API. And zero data retention no longer covers every model: the newest frontier release requires a 30-day retention window, and an organization configured for ZDR gets a hard error when it calls that model, so choosing the API with ZDR also means choosing the model. The capability ceiling post has the detail.
Enterprise is the plan where the organization, and no single user, owns the configuration. $20 a seat a month on annual billing, with usage billed on top at API rates and spend limits an admin sets per user and for the whole org. Self-serve starts at 20 seats, sales-assisted at 50. What the seat buys: audit logs and a Compliance API, custom data retention controls, SCIM provisioning and role-based access, IP allowlisting, and a HIPAA-ready configuration with a BAA the organization owner accepts from settings. An owner adds a connector once and every member signs in as themselves, which is how Clio's Vincent, CoCounsel and the official MyCase connector all install on it now.
When it fits: the firm needs those controls on paper, for a client contract, an insurer, or a partner who wants to see who asked what about which matter. A four-attorney firm we wrote about in October chose it for exactly that and started with four pilot users. When it doesn't: a four to fifteen person firm that wants zero data retention and nothing else. ZDR is an agreement on the API, and Enterprise chat keeps a retention floor on conversations, so buying 20 seats to get it would buy the wrong thing.
Individual
Pro and Max
One login, one person. Pro to find out whether Claude helps, Max once the documents are long. Training is a setting you turn off yourself; retention is Anthropic's default.
training off: opt-out retention control audit log 1 seat
The firm
Team
Two to 150 seats, Standard and Premium mixed. No training on your content by default, single sign-on, an admin who decides which connectors the firm can use.
training off by default retention control audit log 2 to 150 seats
Workflows
The API
No seats, pay per token. Zero data retention by agreement at the organization level, the largest context window, and a connector you control between Claude and the practice system.
training off by default ZDR by agreement audit log: yours no seats
Controls
Enterprise
Seat fee plus usage at API rates. Audit logs, a Compliance API, retention controls you set, SCIM, IP allowlisting, a BAA from settings. From 20 seats self-serve, 50 through sales.
training off by default custom retention audit log 20 seats and up
For a law firm the plan question is a confidentiality question. Two things are easy to get wrong. Training-off is a different property from retention: Max with training off is good, and for privileged content the standard to aim for is terms at the organization level, no training by default on Team and Enterprise, zero data retention on the API for anything that runs on its own. The zero data retention breakdown covers why the mechanics differ across plans. And access is the exposure, with logging a separate question: the risk with consumer AI is that a model provider processed privileged material at all. A 2026 federal ruling addressed exactly this; see our read of the privilege decision and the three privilege tiers for the build-versus-buy choices that follow.
One nuance for litigation firms: whether HIPAA applies depends on who your client is. If you represent an insurer or insured that is itself a covered entity, your firm is usually a HIPAA business associate, and a business associate agreement plus safeguarded data handling are in scope; on Claude that means Enterprise, where the BAA lives. If you represent a defendant that is not a covered entity and simply received records in discovery under a protective order, a BAA may not be required. Confirm your status with counsel. Either way, training excluded in writing, and zero data retention for anything automated, is the standard worth holding to for privileged and sensitive material.
Update, September 2026: Weighing ChatGPT instead? OpenAI's legal edition, Astra for Law (September 17, 2026), isn't part of any ChatGPT plan you can buy. It's limited to selected US firms. Astra for Law explained has the access rules, and the three-way comparison of the legal editions covers data terms by plan.
Count the people first, then the material. One person testing goes on Pro and moves to Max the week the documents get long. Two or more people go on Team, with Premium seats for the drafters, and the admin switches on only the connectors the firm has approved. Anything that should run on a schedule, or any privileged material at volume, goes on the API under a zero data retention agreement, through a connector the firm controls. Enterprise comes in when someone needs to see the log, set the retention, or sign a BAA, and the seat count is there. Most firms we meet end up on two of these at once: Team for the people, the API for the jobs.
Several attorneys on one plan
One connector in the firm's own cloud, every attorney signed in as themselves, one audit log. The plan decision and the connector decision are the same decision.
We build legal AI integrations for regulated firms, and the plan question is the first thing we settle on a call, because it sets the data path for everything after it. If you want to run it yourself, our Clio connector is open source and free, on any plan. If you want it installed and the first workflow running, guided setup does that on a call, scoped per firm. If several attorneys will use Claude with the practice system, the per-laptop install becomes the bottleneck: every machine is its own setup and there is no firm-level record of what AI touched. Firm-wide deployment puts one connector in a place your firm controls, every attorney signed in as themselves, one audit log. How much visibility each plan gives you on its own, and what the connector adds, is in the usage visibility post.
It follows from how many people use it and what they feed it. One attorney finding out whether Claude helps: Pro. One person reviewing documents all day: Max. Two or more people at the firm: Team, which starts at two seats, keeps your content out of model training by default and gives an admin control over connectors. A workflow that runs without a person, or privileged material at volume: the API under a zero data retention agreement. Audit logs, retention controls you set, a BAA and provisioning: Enterprise, from 20 seats self-serve or 50 through sales.
Pro has the lowest usage allowance of the paid plans, measured in rolling five-hour windows plus a weekly cap, and long legal documents consume it fast. Two medical records or one deposition in a session can hit the limit. Anthropic now sells usage credits for that moment, which is the point at which Max, with five or twenty times Pro's usage, gets cheaper for a heavy user.
For everyday work with a written policy, usually yes. Team does no model training on your content by default, includes single sign-on, and lets an admin decide which remote and local connectors the firm can use. It has no audit logs, no custom retention controls and no zero data retention: a conversation stays in the product until someone deletes it, and a deleted one leaves back-end storage within 30 days. If the firm needs a log of what AI touched, the connector that reads the practice system has to keep that log itself.
Max lets one person turn off model training, and that is worth doing on day one. It says nothing about retention: the consumer plans keep conversations under Anthropic's default policy, with no retention controls and no zero data retention. For internal, non-privileged work it is fine. For privileged content the firm wants the terms in writing at the organization level, which means Team or Enterprise for chat, or the API under a zero data retention agreement for anything automated. One more thing to know: zero data retention no longer covers every model. The newest frontier release requires a 30-day retention window, and an organization configured for zero retention gets a hard error when it calls that model, so picking the API with ZDR also means picking the model.
Max for one person working in the Claude app. The API for anything that should run without a person pasting: the largest context window for long documents, pay-per-token pricing, zero data retention by agreement at the organization level, and a connection to the practice management system through a connector the firm controls. Most firm-wide builds run on the API for those reasons, with prompt caching and spend limits to keep the per-token bill flat.
Only if the firm needs what the seat buys: audit logs and a Compliance API, custom data retention controls, SCIM provisioning and role-based access, IP allowlisting, and a HIPAA-ready configuration with a BAA the organization owner accepts from settings. It costs $20 a seat a month on annual billing plus usage at API rates, with a 20-seat minimum self-serve and 50 through sales. A four to fifteen person firm that wants zero data retention and nothing else is buying the wrong thing: ZDR is an API agreement, and Enterprise chat keeps a retention floor on conversations.
Not sure which plan your firm needs?
Thirty minutes, your setup, literally: who uses Claude today, on which plan, what they paste into it. You leave with the plan, the connector choice and the data path on one page.
NEWSLETTER
[ WRITTEN BY ]
Saša Sladić
Co-Founder & CEO
Co-Founder and CEO at Oktopeak. Works with founders in legal, healthcare and fintech to get stalled, broken and inherited products into production.
[ LEGAL TECH ]
LEGAL TECH
Jun 3, 2026 · 11 min read
ZDR is not available on Claude Team or Pro chat. It is available on the Anthropic API at the organization level, which a small firm can obtain. Here is the real difference between the chat tiers and the API, why a Canadian or healthcare firm hits a residency wall, and the architecture we use to keep privileged data off retention.
Read ArticleLEGAL TECH
Oct 7, 2026 · 8 min read
Projects, per-user connectors, tool approval, a connector in your own cloud. What each isolates for a four-attorney firm on Enterprise, and what does not.
Read ArticleLEGAL AI
Jul 29, 2026 · 8 min read
Clio Work is Clio's AI workspace with Vincent inside it. Claude writes better and sees none of your Clio. The open-source Clio MCP connector puts Claude on live matters, with a local log.
Read Article[ GET STARTED ]
30-minute call. No pitch deck. Just an honest conversation about your project.
