Claude Code costs for teams: what it costs, how to see it, and where to cap it

Claude Code is Anthropic's coding tool. A developer types a request in a terminal, the text window where commands are typed, and an AI model reads files, runs commands and edits code. Every part of that work is counted in tokens, pieces of text the model reads or writes, which Anthropic's pricing page, read on 4 October 2026, puts at 4 characters or 0.75 of an English word each [3]. What a team pays for Claude Code is the cost of the tokens its developers' sessions use, or the seats that include an allowance of them.

This guide is for the person who approves that spend and does not use the tool every day. It takes Anthropic's documentation on managing costs for an organisation, read on 4 October 2026, as its only source of fact, and it explains every product term in plain words as it goes. The order below is the order in which a manager meets the questions: what it costs, how the company pays, where the figure appears, where it is capped, what the limit messages mean, what an administrator can enforce, and how to run the first month.

What it costs, by Anthropic's figures

Anthropic's documentation gives three figures, all of them Anthropic's own statements about its own product across what it calls enterprise deployments: an average of $13 per developer per active day, $150 to $250 per developer per month, and costs below $30 per active day for 90 percent of users [1]. In the same paragraph Anthropic says per-developer costs vary widely with model selection, codebase size and usage patterns such as running several instances or automation, and it advises starting with a small pilot group and using the tracking tools to establish a baseline before a wider rollout [1]. The page does not define an active day or say how many of them a month the monthly figure assumes.

The figure moves with what the developer does. Anthropic names long sessions that were never cleared, and Opus left as the default model, as the usual causes of unexpectedly high spend on an API or cloud-provider plan [1]. On Anthropic's pricing page, Claude Opus 5.5 is listed at $4 per million input tokens and $20 per million output tokens, and Claude Sonnet 5.5 at $2 and $10, which is why the default model is the first thing to check [3]. Anthropic also says to budget more for a coding seat than a chat seat, because one debugging session can consume more than a day of chat [1]. The page on what Claude Code costs per developer sets out each figure, everything that moves it, and how to run a pilot that produces your own.

The three ways a company pays

Anthropic's documentation says which controls an organisation has depends on how it accesses Claude Code, and it names three setups [1]. On a Claude for Teams or Enterprise plan, each member has a seat, a paid place on the plan, and each seat has an allowance: an amount of usage included in the price, which resets on a rolling five-hour window and a weekly window, is shared with Claude chat and Cowork, and is sized by the seat tier, Standard or Premium [1]. Anthropic says usage inside that allowance is not metered in dollars; dollar figures appear only for usage credits, the paid usage a member draws on past the allowance once an administrator turns them on [1]. On the Claude Console, Anthropic's account for programs that call its models, usage is billed per token to the organisation, and a workspace called Claude Code, a named section of the account, is created automatically the first time Claude Code authenticates [1]. On a cloud provider, Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry, usage is billed per token to the cloud account [1]. If an organisation mixes sign-in methods, each developer is metered according to the one they authenticated with [1].

Setup (Anthropic's three)	How usage is metered	Where spend appears	Where it is capped
Claude for Teams or Enterprise	Each member's seat allowance; usage credits in dollars past it [1]	Spend report in organisation analytics, updated daily, once usage credits are on [1]	Spend limits in admin settings at organisation, group or member level [1]
Claude Console	Per token, to the organisation [1]	Console usage page for billing; the dashboard for per-member spend [1][2]	Workspace spend limit on the Claude Code workspace [1]
Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry	Per token, to the cloud account [1]	The cloud billing console [1]	The cloud's budget controls [1]

The page on subscription seats or pay per token compares the three and says which to pick.

Where to see spend, and how to get a number per developer

Each setup has its own screen. On a Teams or Enterprise plan, the spend report in organisation analytics shows estimated spend per user and per model, with CSV export, updated daily, and appears once usage credits are turned on; a second screen, the analytics dashboard, shows daily active users and sessions, and Anthropic says Admins and Owners can view it [1][2]. On the Claude Console, Anthropic names the usage page as the place for authoritative billing and the dashboard at platform.claude.com/claude-code for daily costs and a per-member table, with a note that the dashboard's spend figures are estimates for analytics purposes [1][2]. On a cloud provider, spend is in the cloud billing console, because Claude Code sends no metrics from a cloud back to Anthropic, and Anthropic's dashboards and its Claude Code Analytics API do not cover that usage [1].

A developer's own /usage screen is a local estimate: Anthropic says Claude Code computes the dollar figure locally from token counts at list price, that it is an estimate, and that the Console usage page is the authoritative source [1]. An organisation on contracted rates can make the figure match its bill with the modelPricing managed setting, which requires Claude Code version 2.1.242 or later, is read from managed settings only, and changes what Claude Code reports while Anthropic bills at the contracted rate regardless [1][7]. The page on where to see Claude Code spend lists every screen and how often each is updated.

A number per developer comes from a different source on each setup. On the Teams plan it is the spend report CSV, a file of rows that opens in a spreadsheet [1]. On the Enterprise plan the Enterprise Analytics API, an interface a script calls to fetch data, returns per-user usage and cost reports across Claude surfaces; a Primary Owner creates a key with the read:analytics scope, and Anthropic says the API is not available on the Teams plan [1][2]. On the Claude Console the dashboard's team insights table shows each member's spend this month, and the Claude Code Analytics API returns the same daily per-user metrics with an Admin API key [1][2]. On every setup, OpenTelemetry export, which sends figures from each developer's machine to a monitoring system the company runs, is in Anthropic's words the only option that streams per-user token and cost metrics into your own observability stack in near real time [1]. On a cloud provider, the other route is a gateway, a server between Claude Code and the model that tracks spend per key; Anthropic names LiteLLM as one that several large enterprises reported using, and states that the project is unaffiliated with Anthropic and has not been audited for security [1]. The page on per-developer usage reporting explains each source and the metrics Claude Code exports.

Where to cap it

On a Teams or Enterprise plan, Anthropic names the seat allowance as the default upper limit on spend; to let members continue past it, an administrator turns on usage credits and sets spend limits at the organisation, group or individual member level [1]. On the Claude Console, the cap is a spend limit on the Claude Code workspace [1]. On a cloud provider it is the cloud's own budget controls [1]. An organisation that runs a self-hosted Claude apps gateway can cap each developer there too: Anthropic's gateway documentation says caps are set per organisation, group or user by day, week or month, in whole US cents, that Claude Code warns the developer once usage passes 75 percent and again past 95 percent of their most-consumed cap, and that caps reset on UTC calendar boundaries [8]. For a single scripted run there is a per-run dollar cap, --max-budget-usd, which Anthropic's cost page names and the cap page explains [1]. The page on spend limits for an organisation, a group and a person gives, for each setup, where the cap is set and what the developer sees at it.

Rate limits, and the messages developers bring you

A rate limit is a different control from a spend limit: it is the most work an organisation may send to the model in one minute, measured in tokens per minute and requests per minute. For the Claude Console setup, Anthropic gives a table of recommended figures per user by team size, from 200,000 to 300,000 tokens per minute per user for a team of 1 to 5 users down to 10,000 to 15,000 above 500 users, and a worked example of 200 users at 20,000 each, which it states as 4 million tokens per minute in total [1]. The per-user figure falls as the team grows because, in Anthropic's words, fewer users tend to use Claude Code concurrently in larger organisations, and the limits apply at the organisation level, so one developer can temporarily use more than their share [1]. For organisations with custom rate limits, Claude Code traffic in the Claude Code workspace counts toward the organisation's overall API rate limits, and a workspace rate limit on that workspace's Limits page caps Claude Code's share to protect other production workloads [1]. The page on rate limits per user by team size reproduces the table and adds the organisation-wide total for each row.

When a developer is stopped, the message names which limit it was, and Anthropic's documentation lists five situations that mean different things [1]. You've hit your session limit or You've hit your weekly limit is a seat allowance window, shared across all models, so switching models with /model does not restore access; the message shows the reset time [1][5]. You've hit your Opus limit or You've hit your Sonnet limit applies to one model family, and switching outside it with /model keeps the developer working [1]. You've hit your individual spend limit, org's monthly spend limit or team's shared budget means the request would be billed to usage credits and those have reached a limit the administrator set, which the administrator raises at Organization settings > Usage [1][5]. A lowercase spend limit reached comes from a self-hosted gateway [5]. A context or auto-compact warning is a different thing from a usage limit: the conversation is close to the point where Claude Code summarises older history, which Anthropic calls compaction, to free space in the context window, the set of text the model reads on every request [1]. On Claude Code version 2.1.234 or later, an interactive session on a subscription waits in place and continues the task after a limit resets, and the autoContinueAtUsageLimit setting, read from user settings, --settings and managed settings, decides whether that wait starts on its own [6][7]. The page on what each limit message means gives each message, who acts and the action.

What an administrator can enforce

Managed settings are settings an organisation deploys to every developer's machine, and Anthropic says Claude Code applies them above every other level, so no user, project, local or command-line value overrides them, apart from a few security-sensitive exceptions [4]. They arrive in one of four ways: server-managed settings from the claude.ai admin console or a self-hosted gateway, an MDM or operating-system policy, a managed-settings.json file in a system directory, or the user-writable Windows HKCU registry, and /status on a developer's machine shows which source is in effect [4]. The cost-related keys Anthropic names are modelPricing, which is managed-only; the model key and the organization default model, both of which Anthropic describes as defaults, with availableModels as the restriction; maxEffortLevel, which caps the effort level, the amount of reasoning the model does before answering; autoContinueAtUsageLimit; the prompt cache lifetime settings promptCacheTtl and subagentPromptCacheTtl; the environment variable (a named value the operating system passes to a program when it starts) that turns agent teams on, which Anthropic says use what it puts at 7 times more tokens than a standard session when teammates run in plan mode and are off by default; and crossSessionInbound, which decides whether an idle session receives messages from the developer's other sessions as new turns [1][7]. The page on settings an administrator can enforce gives each key's scope and version and says which should stay a team choice.

A first-month plan

Reveneau's recommendation, from Anthropic's advice to pilot and baseline first, is the order below. The durations are Reveneau's; Anthropic gives none.

Week	What to do	Page
Before the start	Choose the setup, because every later control depends on it [1]	Subscription seats or pay per token
Before the start	Set the cap for that setup, and on the Claude Console request a rate limit from Anthropic's table [1]	Spend limits and rate limits
Weeks one and two	Run a small pilot group across at least one full weekly allowance window [1]	What Claude Code costs per developer
Week three	Read the spend screen for the setup, count active days, and divide to get a cost per active day; reconcile against the invoice [1][2]	Where to see spend and per-developer reporting
Week three	Deploy the managed settings the organisation wants, starting with modelPricing if on contracted rates [1][4]	Settings an administrator can enforce
Week four	Compare the measured figure with Anthropic's $13 and $30, decide the budget, name a monthly reviewer, and widen the rollout	Budget worksheet

The Claude Code budget worksheet is a printable form for the whole plan. It also shows Anthropic's three figures multiplied out for 10, 50 and 200 developers, labelled as the page's own arithmetic and as a forecast for no team, because the figures are averages across other companies' deployments and Anthropic does not define an active day.

For the developers

The pages above are for the person who pays. The sister guides are for the people whose habits set the figure. How to reduce Claude Code token usage covers what a developer does in a session: clearing between tasks, choosing the model and effort level, keeping the project notes short, and the other habits Anthropic lists. How the Claude Code prompt cache works covers the cache, the store of request text the model service has already processed and reads back at a lower price, and the actions that reset it. A manager who sends a developer one page should send the first; a manager who wants to know why the first message after lunch costs more should read the second.

Reveneau is an AI software development consultancy. All of its code is written by AI, and every change must pass an eval suite, a set of automated tests written from the specification, before release, so token use is a running cost of every Reveneau build. Reveneau is independent of Anthropic: every figure, screen name, setting and version number in this guide is Anthropic's own statement in documentation read on 4 October 2026, and Reveneau's own contributions are the recommendations, each labelled as such. If you want help choosing a setup, running a pilot or writing the policy, contact Reveneau.