Is Claude suited to summarizing a hundred-page contract or report?

Yes, that's its strong ground: the latest models (Sonnet 5, Opus 5) accept up to a million tokens of context, plenty to fit a long document into a single conversation without splitting it by hand.

Do my documents train Anthropic's models?

No, by default, whether you're on a free, Pro or professional account. Training on your exchanges requires that you've turned it on yourself in the privacy settings, or that you've sent feedback via the thumbs-up/thumbs-down button.

How much does Claude cost for a team of ten people?

The Team offer starts at $20 per user per month on an annual commitment ($25 month to month) for a standard seat, and $100 per user per month for a premium seat. Everything is billed in dollars, before tax.

Where is my data hosted?

On servers located in the United States, or in other countries outside the European Economic Area and the United Kingdom, according to Anthropic's privacy policy. No European hosting is announced as of today.

What's the difference with ChatGPT or Mistral Le Chat?

The most verifiable difference lies in the context window and the default training policy: Claude trains on no conversation without an explicit user action, whatever the tier. For the other two, this distinction between consumer and professional offer is either fuzzier or reversed.

Claude: our review

Claude is Anthropic's conversational assistant, built around a large context window: it reads a whole file rather than excerpts and holds a style instruction across an entire conversation. Companies use it for summarizing long documents and framed writing.

Data in the United StatesFree version, then paid·from $17/month

Go to Claude's website

What we take away from it

What counts in its favor

A context window of up to a million tokens on the latest models (Sonnet 5, Opus 5): a report running several hundred pages fits in a single conversation, with no manual splitting.
No conversation, consumer or professional, trains the models by default. You have to turn it on yourself in the settings, and exchanges in incognito browsing are excluded regardless of that setting.
An Enterprise offer compliant with SOC 2 and ISO 27001, GDPR and CCPA, with retention controls and audit logs — a compliance file already ready for a procurement department.
The Projects feature switches automatically to retrieval-augmented generation (RAG) when a folder exceeds the context window, without the user having to split their files.

What counts against it

Billed in dollars, with no euro tier: the listed Pro price of $17/month assumes an annual commitment paid upfront ($200), otherwise it rises to $20 a month.
Data travels through servers located in the United States or outside the European Economic Area: no European hosting is announced, unlike tools that make it a selling point.
The health offer is only "HIPAA-ready," not HIPAA-certified: a health facility must check what that actually covers before relying on it.

The right choice for an organization that handles long documents and wants text held consistent from start to finish, provided it accepts dollar billing and servers outside the European Union. Rule it out if strictly European hosting is a contractual requirement.

The facts

First paid tier: $17/month — Pro, billed annually
Data hosting: Data in the United States · source
Categories: AI assistants,Writing,Coding assistants
Data verified on: 30 July 2026

What happens to your data

Is your data used to train their models?: No
Subprocessors: By default, neither consumer exchanges (Free, Pro, Max) nor professional offers (Claude for Work, API) train the models. Two documented exceptions: a user who turns on model improvement in their settings, and feedback sent via the thumbs-up/thumbs-down button, kept for up to five years and de-identified before use. Conversations in incognito browsing escape both cases.
Retention: A deleted conversation leaves the servers within 30 days. Model improvement data, when the user has turned it on, is kept for up to 5 years. An exchange flagged for a usage-policy violation can be kept for up to 2 years, and the associated safety scores for up to 7 years.
Data processing agreement: Document published by the publisher

Claude in detail

Claude is Anthropic's conversational assistant. It stands out first for its ability to read a whole document — a contract, a report, a set of contracts — rather than in excerpts, and to hold a style instruction across an entire exchange without losing it along the way.

What it actually does

The most common use among our clients fits in one sentence: pasting a long document and asking for a summary, a proofread or an answer built on that specific basis. The latest models (Sonnet 5, Opus 5) accept up to a million tokens of context, easily enough to fit several hundred pages in a single conversation. Beyond that, the Projects feature takes over: it organises a workspace by folder, with its own documents and instructions, and switches automatically to retrieval-augmented generation (RAG) once content exceeds the context window, without the user having to split files by hand.

For writing, an instruction set once — tone, format, angle — applies to a whole series of texts produced within the same project. That's what sets structured writing use apart from a simple one-off exchange.

Why it interests us

The first question an owner asks is about training: will their documents be used to improve a competitor's model? On this point, Anthropic publishes a clear, dated answer: by default, no conversation is used for training, whether the account is Free, Pro, or a business offer. You have to turn it on yourself in the privacy settings, and exchanges in incognito browsing escape it regardless. That's a more protective default policy than what some competitors in this comparison show, and it applies identically to both audiences — a rare point we cover in our AI agents page.

Where it falls short

Two reservations to know before rolling the tool out across a team.

The first is hosting. Anthropic's privacy policy is explicit: data passes through servers located in the United States, or in other countries outside the European Economic Area and the United Kingdom. No European hosting option is announced to date. For an organisation bound by a strict EU-hosting contract clause, that's a dealbreaker.

The second is billing, in dollars with no euro tier. The advertised Pro rate of $17 a month assumes an annual commitment paid upfront; with no commitment, it climbs to $20. On the Team tiers, the gap between monthly and annual billing works the same way.

One last nuance, for the healthcare sector: the offer is described as "HIPAA-ready," which isn't the same as a full HIPAA certification. It's worth checking precisely what that label covers before relying on it for medical use.

Our recommendation

If your work revolves around long documents and dollar billing isn't a problem, Claude keeps a real promise on context reading and on data handling by default. If European hosting is a non-negotiable clause, or if you need a complete healthcare certification rather than a "ready" offer, look elsewhere — see our comparison of AI tools for the healthcare sector for that last case.

Who it's for, and who it isn't

Suited for

Firms and departments that handle long files — contracts, reports, briefs — and want a faithful summary without splitting the document into pieces
Writing teams that use a fixed project instruction to enforce a consistent tone and format across a whole series of texts
A company that has already settled the question of hosting outside the EU and doesn't see it as an obstacle

Not a fit

An organization bound by a contractual or regulatory requirement of strictly European hosting: Claude doesn't offer it
A team that wants to pay in euros with no exchange-rate risk: billing stays in dollars, whatever the tier
A health facility that needs a full HIPAA certification rather than a mere "ready for" offer

Frequently asked questions

Is Claude suited to summarizing a hundred-page contract or report?: Yes, that's its strong ground: the latest models (Sonnet 5, Opus 5) accept up to a million tokens of context, plenty to fit a long document into a single conversation without splitting it by hand.
Do my documents train Anthropic's models?: No, by default, whether you're on a free, Pro or professional account. Training on your exchanges requires that you've turned it on yourself in the privacy settings, or that you've sent feedback via the thumbs-up/thumbs-down button.
How much does Claude cost for a team of ten people?: The Team offer starts at $20 per user per month on an annual commitment ($25 month to month) for a standard seat, and $100 per user per month for a premium seat. Everything is billed in dollars, before tax.
Where is my data hosted?: On servers located in the United States, or in other countries outside the European Economic Area and the United Kingdom, according to Anthropic's privacy policy. No European hosting is announced as of today.
What's the difference with ChatGPT or Mistral Le Chat?: The most verifiable difference lies in the context window and the default training policy: Claude trains on no conversation without an explicit user action, whatever the tier. For the other two, this distinction between consumer and professional offer is either fuzzier or reversed.