Claude Code pricing, plans and what it really costs per person

Claude Code ships with every paid Claude plan, with no separate licence. The entry ticket starts at around 19 euros a month on the Pro plan. It rises to roughly 95 and then 190 euros on the two Max tiers. The Team plan includes it from its cheapest seat. The other route is to take no subscription at all and pay per token through an API key.

This article lays out the full grid as recorded in September 2026 on Anthropic's pricing page. It then covers what one person actually consumes each month, the two ceilings that frame usage and the levers that bring the bill down. For Claude subscriptions in general, outside Claude Code, our article on Claude pricing covers the topic.

How much does Claude Code cost in September 2026?

Claude Code has no price of its own. It is paid for in four ways, and the choice mostly depends on how many people are involved and how much administrative control you need. A freelancer takes an individual plan. A team buys seats. An IT department that wants audit logs takes Enterprise. A developer automating a one-off job goes through an API key.

The Pro plan can also be paid yearly, at around 16 euros a month billed in one go. Anthropic's page states that the amounts shown exclude tax, so a French consumer adds VAT on top. These prices are published in dollars, and the euro figures above are orders of magnitude.

Claude Code ships with the cheapest Team seat

This is the most widespread mistake on the subject, and an expensive one. Most published comparisons claim a team must buy Premium seats at around 95 euros to get Claude Code. Anthropic's documentation says the exact opposite, « Claude Code is included with every Team plan seat ». The Premium seat does not unlock access, it widens the allowance.

The gap matters for a small company. A team of ten that follows those comparisons budgets around 11,400 euros a year. The same team on Standard seats pays around 2,280 euros a year for the same tool. You can mix both seat types in one organisation and move only the heaviest users to Premium.

Claude Code Enterprise pricing is no longer quote-only

For a long time the Enterprise plan pointed you to a salesperson. That is no longer the case. Anthropic now publishes a seat price, around 19 euros per seat per month on an annual commitment, plus usage billed at API rates. The logic differs completely from Team, where the allowance is baked into the seat.

In practice an IT department pays little for the base and a lot for actual usage. That base buys control, single sign-on, identity management, audit logs, network restrictions and spend limits per person or per group. Across a fleet of developers, that control is what justifies the move, not the entry price.

What one person really consumes each month

Anthropic publishes a figure almost nobody quotes, and it is the most useful one for building a budget. The figure comes from its own enterprise deployments. the Claude Code documentation measures around 12 euros per person per active day. The monthly range runs from 140 to 240 euros per person. Nine users out of ten stay under 28 euros a day.

That benchmark settles the flat-rate versus usage question. Someone who codes every day consumes more, at API rates, than the price of a Max plan at around 190 euros a month. The flat rate pays off as soon as usage becomes daily. Conversely, someone who opens Claude Code twice a month will pay less per token.

I put it differently in my newsletter edition on Claude Code, a single coding session eats the equivalent of about twenty ordinary conversations. On a Pro plan the ceiling arrives fast. That is the moment the next tier becomes a real question.

Two ceilings, and switching models unlocks neither

A Claude subscription does not buy unlimited volume. It opens two counters that run at the same time, a rolling five-hour window and a weekly cap. Both are shared with Claude chat, so a day of conversations eats into the evening's coding budget.

When one of those ceilings hits, switching to a cheaper model changes nothing, because the counter covers every model. Two options remain. Wait for the reset, whose time is displayed. Or turn on usage credits, which let you keep working at API rates, under a monthly spend limit you set yourself.

Should you use the API rather than a subscription?

An API key removes the subscription and bills every token. It is the right option for automated use, for a one-off job, or for an organisation that wants all spend to flow through a single console. The model you pick then decides the price, in a range of one to four between the two ends of the catalogue.

One subtlety invalidates the naive reading of that grid. The most recent models use a different tokeniser, which produces around 30 percent more tokens for the same text. A lower headline price therefore does not mechanically produce a lower bill. Compare complete tasks, not prices per million tokens.

In that video I walk through the whole setup. The budgeting point is that an API key requires no commitment. You top up a few euros and you consume. To test the tool before committing a team, it is the least risky route.

What a Claude Code subscription does not replace

One budgeting mistake keeps coming back in companies, cancelling other tools on the assumption that Claude Code absorbs them. I wrote about it on LinkedIn in February 2026. A coding agent reproduces the logic of an automation scenario very well. It does not replace the infrastructure that runs it.

An honest calculation therefore adds up the subscription, the hosting and the platforms you keep. When Tandem rebuilt this site, the tooling bill came down to the Claude subscription plus around twenty euros a month of hosting. A website builder subscription disappeared along the way, which offset the spend. That reasoning is done line by line.

The detail of that rebuild and its real cost is told in our write-up on rebuilding the site with Claude.

Five levers that bring the bill down

The headline price matters less than the way people work. On our engagements, consumption varies threefold between two people doing the same job on the same plan. Five habits explain most of that gap.

1. Start a clean session between topics. Stale context is paid for again on every following message.
2. Match the model to the task. Sonnet handles most coding, Opus is kept for architecture decisions.
3. Keep the instructions file short. Anthropic recommends staying under two hundred lines, because that file loads on every session.
4. Turn off unused connectors. Every connected server takes up context before the first question is even asked.
5. Keep agent teams small. Several agents running in parallel consume roughly seven times more tokens than a single session.

The /usage command shows recent consumption and names what weighs, skills, subagents, connectors. It is the first reflex to install in a team, before any discussion about plan tiers. The rest of the good practices are covered in our Claude Code guide for non-developers.

Which Claude Code plan should you choose?

For regular personal use, take Pro at around 19 euros a month and watch how fast you hit the ceiling. If you hit it every week, move to Max. The first tier at around 95 euros is enough in the vast majority of cases, and the second is justified when you run several agents in parallel.

For a team, start on Standard Team seats at around 19 euros per person and move only the heaviest users to Premium. For an organisation with traceability obligations, Enterprise is the only plan that provides audit logs and per-person spend limits. And keep the API key for what it does best, automated jobs and commitment-free testing.

The real budget risk is not the entry price, it sits in working habits. A team that has never opened /usage will pay more for an identical result. That is exactly what we work on during our AI literacy sessions and in our Claude Code training programme.