Claude Code
Anthropic PBC ·claude.com/claude-code· As of: 2026-07-30
Our verdict
Coding agents went from exception to expectation within two years. Working without one costs speed. Working with one does not make you distinguishable, because the same shortcut is open to everyone. The edge only comes from the context you give the agent: your own data, your own rules, your own review chains.
Who it fits
Solo
suitable
Seventeen dollars a month billed annually, twenty billed monthly. For one person who builds, that is the cheapest line item in the whole toolbox.
Mid-market
suitable
Twenty dollars per seat per month billed annually, twenty-five monthly. The question is not the price but who signs off on the output.
Enterprise
suitable with caveats
In the enterprise plan, usage at API rates is added to the seat price. The real hurdle is not money but approval: may code leave the building, and is the process logged.
Buy, switch or build
Off-the-shelf
No established off-the-shelf product
The category is younger than three years. There is no grown standard product to buy instead, so the comparison runs between AI-native vendors and building it yourself.
AI-native
Editor-bound assistance
Assistance inside the editor stays closer to familiar habits and is easier to introduce. An agent working in the directory can take on more connected steps, but it demands a review chain that was not needed before.
Build it yourself
Your own agent scaffold on model APIs
Worth it when the workflow is narrow and repeated and the rules come from inside the house. It regularly fails on maintenance: models, tool interfaces and security requirements move faster than a self-built scaffold gets maintained.
Our advice
- Solo: Take it. The cheapest lever available.
- Mid-market: Take it, but decide at the same time who signs off on output and what never gets committed unchecked.
- Enterprise: Settle approval before rollout: training exclusion, processing location and logging, in writing.
Who owns what gets built
files in your own repository · portable
The output is files in your own directory and your own version control. If the tool goes away the code stays, runnable and maintainable by others. That is the difference from a builder that the application lives inside.
Who is behind it
Staying power: Funded
Anthropic only started in 2021 and runs on unusually large venture capital: two rounds in 2026 alone, at 30 and 65 billion US dollars, most recently at a 965 billion US dollar valuation. That valuation is the price agreed in a private round, not a market quote. The scale does not make the company established, because audited accounts are missing, profitability is not proven, every revenue figure is the company's own statement, and staying power depends on further capital and on very large compute commitments. On the other side of the ledger sit a self-reported 47 billion US dollar revenue run rate, a Long-Term Benefit Trust that has held the majority of the board since April 2026, and the confidential S-1 submission, which makes a listing possible without committing to one. The ownership picture is only half clear: Google's stake is documented in court filings and contractually capped, while Amazon's stake is undisclosed and merely estimated. That carries the classification funded, not established.
- Series H: 65bn USD at a 965bn USD valuation: USD 65bn, led by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital, post-money USD 965bn. The 65bn includes USD 15bn of investment already committed by hyperscalers, USD 5bn of it from Amazon, so the amount of genuinely new third-party money is correspondingly smaller. The valuation is the price agreed in this private round, not a market quote. · Source · As of 2026-05-28
- Series G: 30bn USD at a 380bn USD valuation: USD 30bn, led by GIC and Coatue, with D. E. Shaw Ventures, Dragoneer, Founders Fund, ICONIQ and MGX as co-leads · Source · As of 2026-02-12
- Revenue run rate of 47bn USD, self-reported: The run rate crossed USD 47bn in May 2026, up from USD 14bn at the Series G in February 2026. Both figures come from the company itself. A run rate annualises a point in time and is not an audited annual revenue figure; no external audit is available. · Source · As of 2026-05-28
- Claude Code as a revenue line of its own: Run rate above USD 2.5bn, more than doubled since the beginning of 2026; business subscriptions to Claude Code quadrupled since the start of 2026; more than half of Claude Code revenue comes from enterprise use. All figures are the company's own, dated February 2026, and refer to Claude Code, not to total revenue. · Source · As of 2026-02-12
- Legal form and control structure: Anthropic, PBC, incorporated in Delaware as a public benefit corporation. The Long-Term Benefit Trust holds a share class of its own (Class T) with five financially disinterested trustees and elects board seats that phase in against time and funding milestones, reaching a majority within four years. The arrangement is not untouchable: stockholders can amend it without trustee consent if sufficiently large supermajorities agree. · Source · As of 2023-09-19
- Trust has held the board majority since April 2026: With the appointment of Vas Narasimhan on 14 April 2026, directors appointed by the Long-Term Benefit Trust make up the majority of the seven-member board. · Source · As of 2026-04-14
- Confidential S-1 submission to the SEC: Anthropic, PBC has confidentially submitted a draft registration statement on Form S-1 to the US Securities and Exchange Commission. Share count and price are not set. The submission creates the option of going public once the regulator completes its review; whether and when that happens depends, by the company's own statement, on market conditions and other factors. · Source · As of 2026-06-01
- Ownership held by the strategic investors: Court filings put Google's holding at around 14 percent in straight equity, contractually hard-capped at 15 percent. Amazon does not disclose its stake; estimates based on public filings place it in the mid to high teens percent, held through roughly USD 42.2bn of convertible notes and roughly USD 32bn of non-voting preferred stock. The size of Amazon's stake is therefore an estimate, not a documented figure. · Source · As of 2026-06-04
- Enterprise customer base: More than 500 customers spend over USD 1m a year on Claude, against about a dozen two years earlier. The number of customers spending more than USD 100,000 a year grew sevenfold over the past year. Eight of the Fortune 10 are Claude customers. The figures are the company's own and cover Claude as a whole, not Claude Code alone. · Source · As of 2026-02-12
- Certifications covering the commercial products: ISO 27001:2022, ISO/IEC 42001:2023 (AI management system), SOC 2 Type I and Type II, plus a HIPAA-ready configuration with a business associate agreement available on request, covering Claude for Work and the Anthropic API. The consumer plans Free, Pro and Max are not covered. · Source · As of 2026-03-16
Cost of leaving: Low
The tool holds nothing back. It works on files in your own version control, and those keep running without it.
Regulation and data
| Data processing agreement | yes, published in the trust centerSource · As of 2026-07-30 |
|---|---|
| Storage location | USA; requests may be routed through locations in the US, Europe, Asia and Australia, but storage is in the USFor Claude Enterprise and the Anthropic API the vendor states plainly that data is stored in the US, and that traffic may be routed to selected countries across the US, Europe, Asia and Australia. It makes no EU data residency commitment for the first-party offering. The source document is dated 15 June 2026.Source · As of 2026-07-30 |
| Subprocessors | on request via the trust centerSensitive documents sit behind an access request.Source · As of 2026-07-30 |
| Third-country transfer | yes, transfer to the US; covered by the EU standard contractual clauses, the UK addendum and a Swiss addendum inside the data processing agreementThe data processing addendum incorporates the standard contractual clauses in module two and module three, plus the UK addendum in version B.1.0 and a Swiss addendum. Under the commercial terms the contracting entity for the European Economic Area, Switzerland and the United Kingdom is Anthropic Ireland, Limited. The addendum is effective 24 February 2025.Source · As of 2026-07-30 |
| Training on customer data | no, under commercial terms there is no training on prompts or codeThe Claude Code documentation states for Team, Enterprise, the API and third-party platforms that Anthropic does not train generative models on code or prompts sent to Claude Code under commercial terms. The exception is an explicit opt-in to the Development Partner Program by an organisation admin. The commercial terms carry the same commitment. On the consumer plans Free, Pro and Max it depends instead on an account setting.Source · As of 2026-07-30 |
| Retention and deletion | 30 days as the commercial default; zero data retention only for eligible Enterprise accounts once enabledTeam, Enterprise and API usage carries a 30 day retention period. Zero data retention is not part of the standard Enterprise plan; it is switched on per organisation after an eligibility check. Separately, the client keeps session transcripts locally in plain text, 30 days by default, adjustable via cleanupPeriodDays. Transcripts shared through the feedback commands are kept for five years.Source · As of 2026-07-30 |
| Certifications | SOC 2 Type 2, ISO 27001:2022, ISO/IEC 42001:2023, CSA STAR, HIPAA, NIST 800-171, FedRAMP High for the government offeringISO/IEC 42001 is the standard for AI management systems. It is currently the most solid signal that a vendor runs its AI processes in an orderly way.Source · As of 2026-07-30 |
| EU AI Act, Article 50 | the vendor publishes model documentation and is certified to ISO/IEC 42001The Article 50 transparency duty falls on the deployer. What the vendor supplies is the basis for it, not its fulfilment.Source · As of 2026-07-30 |
| Audit logging | yes on the Enterprise plan: per-event records via the compliance API and a CSV audit export covering 180 days; agent runs themselves only through a self-operated OpenTelemetry collectorThe compliance API gives Enterprise organisations per-event records from the activity feed. The CSV export in organisation settings reaches back 180 days and carries identifiers only, not content. What the agent does inside the directory is emitted by the client as OpenTelemetry events to a collector the customer runs. Keeping an audit-proof record therefore stays the customer's job.Source · As of 2026-07-30 |
What it really costs
Entry price
Included in the Pro (17 dollars a month billed annually, 200 dollars up front, or 20 dollars monthly), Max (from 100 dollars a month), Team (20 dollars per seat annually, 25 monthly) and Enterprise plans.
Source · As of2026-07-30
What gets expensive
In the enterprise plan: seat price plus usage at API rates. Anyone running an agent continuously is no longer paying for the seat but for the usage.
What it displaces
- Hand-written boilerplate
- Small outsourced jobs for self-contained components
Interfaces
- Terminal and version control
- Model APIs
- Tool integration via MCP
At Convios: In use at Convios
In daily use, including for this site. The corresponding review chain sits in the repository and runs before every publication.
Matching methods
Alternatives
- Cursor
- Lovable
Evidence
- 1.Certifications, DPA, model documentation · As of 2026-07-30
- 2.Plans and prices, and which plans include Claude Code · As of 2026-07-30
Last reviewed: 2026-07-30byDr. Oliver Gausmann, Convios GmbH
Details out of date? Let us know.