Quick answers
Is a consumer plan safe for client data? I wouldn't use one for it, because Max and Ultra run under the Consumer Terms, which carry no data processing addendum making Anthropic your processor.
Is Max cheaper than Team? Not per unit, since Max costs the same per five-hour session as a Team seat billed monthly, and what Ultra adds is a higher ceiling with no five-hour limit.
What split would I use? Consumer plans for build work on dummy data, and Team or Enterprise seats for anyone who touches clients' personal data.
No, I wouldn’t put clients’ personal data into a consumer Claude plan. And no, Max isn’t cheaper than Team once you divide price by usage. What Ultra does buy is a higher ceiling for one heavy user than any Team seat, and either consumer plan is fine for build work on dummy data. Team and Enterprise come under Anthropic’s Commercial Terms, with a data processing addendum that makes Anthropic your processor. Max and Ultra don’t, so client personal data belongs on the business plans.
This is my reading of Anthropic’s published terms and UK GDPR as they stood on 5 November 2026. It isn’t legal advice. If your firm handles client data in finance, brokerage or professional services, your data protection lead should check it against your own contracts.
How do the plans compare on price?
These are the plans a 5 to 50 person firm would weigh, as Anthropic listed them on 5 November 2026. US prices, per person per month, before tax.
| Plan | Price | Usage allowance | Terms |
|---|---|---|---|
| Max | $100 | 5x Pro every five hours and every week | Consumer |
| Ultra | $200 | 10x Pro every week, no five-hour limit once verified | Consumer |
| Team Standard | $25, or $20 billed annually | 1.25x Pro per session, plus a weekly limit | Commercial |
| Team Premium | $125, or $100 billed annually | 6.25x Pro per session, plus a weekly limit | Commercial |
| Enterprise | $20 a seat billed annually, 20 seat minimum, plus usage at API rates | Usage cost scales with model and task | Commercial |
Ultra is the plan Anthropic used to call Max 20x, and existing Max 20x subscribers moved to it with no price change. A Team plan needs at least two members and stops at 150. Since Anthropic’s January 2026 Team update, Claude Code comes with every Team seat.
Divide each price in the table by its multiplier, and Max, Team Standard and Team Premium land on the same price per unit of five-hour usage on monthly billing. Billed annually, both Team seats come out a fifth cheaper per unit. So per session the consumer plan isn’t cheaper, and against annual Team billing it’s dearer.
The weekly picture is murkier, because Anthropic doesn’t publish how a Team seat’s weekly limit compares with Pro. Wokey, which sells pay-as-you-go Claude access and so isn’t neutral, measured Max 20x accounts at 1.8 times a Team Premium seat’s weekly limit, for 1.6 times the price. That’s a little more per dollar than Premium billed monthly, and less than Premium billed annually.
Where Ultra pulls ahead is the ceiling. The biggest Team seat stops at 6.25 times Pro per session, while a verified Ultra account has no five-hour limit. For one developer running several agents at once all day, that matters more than the unit price. A feature request on GitHub from April 2026 asks Anthropic for a Max 20x tier on Team, because for power users “6.25x is not enough.” Its author describes their workaround: personal Max 20x accounts for the heavy users and a separate Team org for everyone else, with no shared projects between them. A commenter added in October 2026 that they need Team for its DPA and can’t get 20x usage there.
A couple of extras don’t change these sums. Max, Ultra and Team now include monthly API credits, and on Team they’re pooled up to a monthly cap. They don’t cover interactive Claude Code sessions and expire at the end of each billing cycle. Team admins can also switch on usage credits so members keep working past their seat allowance, and I went through how that billing works in Claude extra usage for teams. Before paying for a bigger plan of either kind, it’s worth cutting what each session burns.
What heavy use is worth at API prices
Against pay-per-token billing, the consumer plans win easily. That matters if you’re weighing Enterprise, where seats come with usage charged at API rates. Anthropic’s own Claude Code cost guide puts the average in enterprise deployments at $150-250 per developer per month, with 90% of users under $30 per active day. Heavy users run far past that.
One developer on Max added up their own Claude Code transcripts for 4 October to 3 November 2026 and priced them at API rates: about $5,000, roughly fifty times what the plan costs. FinOps LLM, which sells cost audits, priced one user’s August 2026 logs at $6,986 at Opus 5 list prices. Wokey’s estimate for Max 20x is about $6,711 a month, measured on 18 real accounts.
My own desktop is an outlier. ccusage priced its Claude Code use at API rates at $80,396 for September 2026 and $83,571 for October. That’s a heavy multi-agent setup, so it isn’t a figure for any one plan. It does show how far past a subscription price API-rate billing can run.
None of these numbers is what Anthropic pays. Martin Alderson made that case in March 2026, writing that “API pricing is not what it costs Anthropic to serve those tokens.” Read the figures as what the same work would cost you on API billing.
Consumer terms and commercial terms
Anthropic’s legal page for Claude Code sets out the split in two lines. Team, Enterprise and API users get the Commercial Terms. Free, Pro, Max and Ultra users get the Consumer Terms. That page also says advertised usage limits on Pro, Max and Ultra “assume ordinary, individual usage” of Claude Code and the Agent SDK.
On Team and Enterprise, your firm is the controller and Anthropic acts as a processor, handling the data only as you instruct. The Commercial Terms pull in Anthropic’s data processing addendum by reference. That DPA, effective 24 February 2025, names Anthropic as your processor, covers UK transfers through the UK Addendum to the EU standard contractual clauses, and gives you 15 days to object when Anthropic adds a subprocessor. The same privacy page says Anthropic doesn’t train on data from its commercial products unless a customer opts into its Development Partner Program.
Anthropic says data on its commercial products is stored in the US. The UK Addendum is what covers sending UK personal data there. If your rules need EU residency, first-party Claude doesn’t offer it.
The Consumer Terms, effective 8 October 2025, are a contract between Anthropic and the individual, covering services Anthropic offers “for individuals.” They don’t ban using a personal plan for work. They also never mention a data processing addendum or a processor role. They do forbid sharing your login or making your account available to anyone else, so one Max account shared round an office breaks them.
Since Anthropic’s consumer terms update of 28 August 2025, Free, Pro and Max users, Claude Code included, choose whether their chats and coding sessions can be used to train Claude. Allow it and retention stretches to five years. Decline and it stays at 30 days. Team, Enterprise and the API were left out of that change. Even with training off, the Consumer Terms let Anthropic train on chats a user gives feedback on or that get flagged for safety review. And the choice sits in each person’s own account settings, so on personal accounts it’s made person by person.
Can a firm keep everyone on Max anyway?
Some firms do run heavy users on personal Max 20x accounts, now Ultra, as that GitHub thread shows, and I can see the pull of the higher ceiling. For client data, that’s a choice to carry a risk.
Start with the law. In UK GDPR, Article 28 says “the controller shall use only processors providing sufficient guarantees,” and that processing by a processor “shall be governed by a contract or other legal act.” Your firm is the controller of its clients’ data. Put that data into a personal Max account and, as I read it, there’s no processor contract between your firm and Anthropic covering it. The Consumer Terms are an agreement between Anthropic and your employee.
Official guidance points the same way. The UK’s National Cyber Security Centre wrote in March 2023 that a query “will be visible to the organisation providing the LLM,” and advised “not to include sensitive information in queries to public LLMs.” Hamburg’s data protection commissioner went further in a November 2023 checklist: staff shouldn’t open chatbot accounts on their own, and if the terms let the provider use data for its own purposes, “you should not transmit any personal data to the AI.” The Consumer Terms reserve that right for feedback and flagged chats, whatever the training setting says.
The mistake these rules guard against keeps happening. The Dutch regulator, the Autoriteit Persoonsgegevens, warned firms in August 2024 after a family doctor’s assistant fed patients’ private information into a ChatGPT-based program and a telecoms worker put a database including addresses into a chatbot. A Dutch accountancy title reported in December 2025 that the regulator had received dozens of reports of sensitive information shared through chatbots in 2024 and 2025. In Australia, a former contractor to the NSW Reconstruction Authority uploaded a spreadsheet of more than 12,000 rows to ChatGPT, and the October 2025 disclosure put the people affected at up to 3,000. The two incidents that name a tool both name ChatGPT.
I haven’t found an ICO fine or reprimand aimed at staff using a consumer AI account, so I won’t overstate the UK enforcement risk today. What a firm keeping client data on Max takes on, as I read it:
- No processor contract with Anthropic for that data, which is what Article 28 expects a controller to have, and nothing to show a client who asks.
- Retention and training decided person by person, with up to five years of retention for anyone who allows training.
- Training on feedback and flagged chats even when training is off.
- A breach of the Consumer Terms the moment an account is shared.
Some firms accept that list. My view is that a Team seat costs no more per session than Max, and a Premium seat allows more per session than Max does, so Ultra’s ceiling is the only thing you’d gain by carrying it.
Split the work by the data it touches
For a 5 to 50 person firm whose staff use Claude Code heavily, this is the split I’d set up.
Build work on dummy data goes on Max or Ultra. If the data is invented, the Consumer Terms aren’t a problem, and for the heaviest users Ultra’s ceiling earns its keep. Make the dummy data properly fake, though. Hamburg’s checklist warns that stripping names and addresses isn’t enough when a person can still be identified from context.
Anyone who opens a client file in Claude goes on a Team seat, Premium for the heavy users. Per session it costs the same as Max on monthly billing and less on annual billing, so the processor agreement costs you nothing extra per unit of usage. What you give up is Ultra’s ceiling. If someone keeps hitting Premium’s limits on client work, that’s a reason to look at Enterprise, which I compared with Team in Claude Team vs Enterprise. Price its API-rate usage against the measurements above first. Finance firms should also read Claude for financial services compliance.
Write the rule down where staff will see it: which plan for which work, and that client data never goes into a personal account. That’s also the first item on Hamburg’s checklist, a written internal rule on which tools can be used and under what conditions.
If anyone on a personal plan has already pasted client data in, find out whether they allowed model training, and have them delete those chats. Anthropic’s consumer retention page says deleted chats leave its back-end storage in no more than 30 days and aren’t used to train future models.
