Spend limits for an organisation, a group and a person: where each Claude Code cap is set

A spend limit for Claude Code, Anthropic's coding tool, can be set at three levels on a Teams or Enterprise plan, the organisation, a group and an individual member, and at matching levels on a self-hosted gateway. On the Claude Console the cap is set on a workspace, and on a cloud provider it is set in the cloud's own budget controls. This page explains where each cap is set, what a developer sees when they reach it, and what the developer and the administrator can each do next, from Anthropic's documentation read on 4 October 2026. It is part of the guide to Claude Code costs for teams.

A token is a piece of text the model reads or writes; all Claude Code usage is counted in tokens. The three ways a company gets Claude Code, and how each is metered, are on the page on subscription seats or pay per token. A rate limit is a different control from a spend limit: it caps how fast requests can be sent, and it has its own page on rate limits per user by team size.

Teams and Enterprise plans: the seat allowance, then usage credits with limits

On a Claude for Teams or Enterprise plan, each member has a seat with an allowance, an amount of usage included in the seat that resets on a rolling five-hour window and a weekly window [1]. Anthropic's documentation states the consequence in one line: the seat allowance is the default upper limit on spend, with nothing else set [1]. With nothing else set, a member who uses it up sees a message such as "You've hit your session limit" or "You've hit your weekly limit", with the reset time, and no usage-credit spend occurs [1].

To let members continue past the allowance, an administrator turns on usage credits, which is paid usage measured in dollars, and "set[s] spend limits at the organization, group, or individual member level" [1]. The controls are in the claude.ai admin console, at Organization settings > Usage (claude.ai/admin-settings/usage) [1]. Three levels mean three kinds of cap:

• Organisation level: one monthly limit for the whole organisation's usage credits.
• Group level: a pooled budget an administrator assigns to a group. Anthropic's error documentation calls this the "team's shared budget" and says it is "a pooled budget an admin assigned to a group you belong to" [4].
• Member level: a limit for one person.

When a developer's request would bill to usage credits and the relevant limit has been reached, the request stops with a message naming whose limit it was. Anthropic's error reference, read on 4 October 2026, lists the variants a Teams or Enterprise member can see [4]:

Message the developer sees	Whose limit	What follows the message
You've hit your individual spend limit	The member's own limit	ask your admin for a higher limit
You've hit your org's monthly spend limit	The organisation's limit	visit claude.ai/admin-settings/usage to raise it
You've hit your team's shared budget	A group's pooled budget	ask your admin to raise it at claude.ai/admin-settings/usage
You've hit your channel's monthly spend limit	One Slack channel's budget	an org owner or channel manager can raise it in the channel's Claude settings

Three details from the same page change how the message reads. The group message does not name the group [4]. Before version 2.1.268, a group's pooled budget produced the "individual spend limit" message instead [4]. And on organisations with usage-based billing, the message says "usage limit" in place of "spend limit", as in "You've hit your individual usage limit" [4]. When one of the plan's usage windows is what ran out, the message also names when that window resets, for example "your session limit resets 3:45pm", and access returns then without anyone raising a limit; before version 2.1.239 the message gave no reset time [4].

What the administrator does: go to Organization settings > Usage and increase the limit the message names, by Anthropic's cost documentation [1]. What the developer can do: wait for a named reset time, or run /usage-credits.

What /usage-credits does, by role

/usage-credits is the command a developer types inside Claude Code to manage usage credits. Anthropic's documentation says it works after signing in with a claude.ai subscription through /login and is unavailable with API key authentication (signing in with a secret key for a Console account instead of a claude.ai login), and that in self-serve Enterprise organisations, Enterprise trials, and Enterprise organisations billed through AWS Marketplace it requires Claude Code version 2.1.248 or later [1]. What it opens depends on the person's role [1]:

• A Pro or Max subscriber is sent to Settings > Usage on claude.ai, where they can turn usage credits on or off and see their balance, this month's spend and their monthly spend limit. On those plans, when the person reaches their own spend limit with credits still available, Claude Code offers to raise or remove the limit from inside the tool [1].
• A Team or Enterprise member with billing access is sent to Organization settings > Usage.
• A Team or Enterprise member without billing access is asked to confirm, and Claude Code then sends a request to the organisation's administrators. Before version 2.1.211 the request went without a confirmation step [1].

Two rules follow for the last group. The confirmation appears only in interactive sessions; in non-interactive mode with the -p flag, and from Remote Control, which is Anthropic's way of continuing a local session from a phone or browser, the command sends no request and tells the developer to run it in an interactive session [1]. And a second run while a request is waiting tells the developer a request has already been sent; after an administrator dismisses it, running the command again sends a new one, which before version 2.1.222 was blocked [1].

A developer can also see their own spend against a limit. On Team and Enterprise plans, /usage shows a usage-credits row with the member's own spend for the current month, measured against any limit that applies to them [1]. A limit that covers the whole organisation is left off the row, and a member with no limit of their own sees their spend with no limit beside it [1]. While usage credits are off for that member, the row is absent; when a limit exists, the row appears as soon as credits are on and shows 0 percent until the first spend [1]. Before version 2.1.236 the row appeared only on Pro and Max plans [1]. A manager who asks a developer to read this row should know that it never shows the organisation-wide cap.

The Claude Console: a spend limit on the Claude Code workspace

On the Claude Console, which is Anthropic's web console for accounts billed per token, spend is organised by workspace, a named compartment of the account. Anthropic's documentation says that when Claude Code first authenticates with a Console account, a workspace called "Claude Code" is created automatically, that it provides central cost tracking for all Claude Code usage, and that no API keys can be created for it [1]. The cap is a workspace spend limit "on total Claude Code spend", set in the Console [1].

Anthropic's error reference adds the reason to set one. The message "Credit balance is too low" means the Console organisation has run out of prepaid credits, and the advice includes adding credits with auto-reload, and setting per-workspace spend caps in the Console so that one project cannot use up the whole organisation balance [4]. The same workspace also has a Limits page where a workspace rate limit can be set, which the page on rate limits per user by team size covers.

For a single scripted run, there is one more cap. Anthropic's command-line reference lists --max-budget-usd, a dollar amount at which a non-interactive run stops; spend from subagents counts toward it, a new subagent fails with "Budget limit reached" once spend reaches the cap, and the cap-enforcement behaviours require Claude Code version 2.1.217 or later [3].

Cloud providers: the cloud's budget controls

On Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry, the Amazon, Google and Microsoft services that offer Claude models inside a company's cloud account, Anthropic's documentation says Claude Code is billed per token to the cloud account "and spend controls live in your cloud provider's billing console" [1]. Anthropic's documentation gives no more detail than that, so the cap is whatever budget control the cloud provider offers on that account. For a per-person cap on a cloud provider, Anthropic names a self-hosted Claude apps gateway [1].

A Claude apps gateway: per-person caps by day, week or month

A Claude apps gateway is Anthropic's self-hosted gateway, a server the company runs between Claude Code and the model provider so that developers sign in to the gateway and the gateway forwards requests with one company credential. Its spend-limits documentation, read on 4 October 2026, says spend limits "cap how much each developer can spend through your Claude apps gateway in a given day, week, or month" [2].

Caps are set through an administrator API on the gateway, an interface that a script or command uses to send instructions to it, at /v1/organizations/spend_limits, with three fields [2]:

• scope.type is user, rbac_group or organization: one developer, one identity-provider group, or the organisation-wide default.
• amount is a whole-number string of US cents, or null for unlimited; "0" blocks every request.
• period is daily, weekly or monthly, and a scope can hold one cap per period, each enforced on its own.

Anthropic's own examples set an organisation-wide default of $500 per month for every developer, written as "50000" cents, and a tighter $100-per-day cap on each member of a contractors group, written as "10000" [2]. A group or organisation cap is a per-seat default that each member gets as their own cap, by the documentation's wording [2]. When several caps apply, the documentation gives the order for each period: a per-user override, then the most restrictive of the person's group caps, then the organisation default, then unlimited [2]. Caps reset on UTC calendar boundaries: daily at 00:00 UTC, weekly on Monday, monthly on the first [2].

At the cap, the gateway returns an error on the next request and blocks the developer until the period resets or an administrator raises the cap [2]. The message names the period and reset time, for example spend limit reached (daily; resets 2026-08-08 00:00 UTC), followed by any text the administrator set in admin.blocked_message; before version 2.1.225 on the gateway server the message read only spend limit reached [2]. Before the cap, Claude Code warns the developer at 75 percent and again at 95 percent of their most-used cap, which needs version 2.1.225 or later on both the developer's machine and the gateway [2]. A Spend limit bar in /usage with the percentage used and the reset time needs version 2.1.251 on the machine and 2.1.225 on the gateway server, and dollar amounts in that bar, such as "$271.40 / $500.00 spent this month", need version 2.1.284 on both sides [2].

Two facts about the gateway's figures. They are estimates computed by the gateway from token counts, and the documentation says to reconcile against the provider's usage reporting for billing; an unrecognised model is priced at $5 per million input tokens and $25 per million output tokens so that it is never free [2]. And if the gateway's database is unreachable, enforcement fails open by default, meaning the request proceeds with a warning in the log; the enforcement.fail_closed_on_error setting blocks requests instead, with the message spend limit unavailable [2].

Every cap in one table

Setup	Where the cap is set	What happens at the cap
Teams or Enterprise, usage credits off	The seat allowance itself, the default upper limit [1]	"You've hit your session limit" or "weekly limit", with the reset time; access returns at the reset [1][4]
Teams or Enterprise, usage credits on	Organisation, group or member spend limits at Organization settings > Usage [1]	A message naming whose limit it was; an administrator raises it there, or the developer waits for a named plan reset [1][4]
Pro or Max, one person	Settings > Usage on claude.ai [1]	With usage credits still available, Claude Code offers to raise or remove the limit from inside the tool [1]
Claude Console	Workspace spend limit on the Claude Code workspace [1]	Spend in that workspace stops at the limit; "Credit balance is too low" when the organisation's prepaid credits run out [1][4]
Claude Console, one scripted run	--max-budget-usd on the command line [3]	The run stops; a new subagent fails with "Budget limit reached" [3]
Cloud provider	The cloud provider's budget controls [1]	Set by the cloud provider
Claude apps gateway	Admin API, per user, group or organisation, per day, week or month [2]	Error on the next request with the period and reset time; warnings at 75 and 95 percent beforehand [2]

Which cap to set first

Reveneau's recommendation, from the facts above. On a Teams or Enterprise plan, decide first whether usage credits should be on at all, because with them off the seat allowance is the upper limit and no usage-credit spend occurs; if they go on, set an organisation-level limit the same day, then member-level limits for the pilot group described on the page on what Claude Code costs per developer. On the Claude Console, set the workspace spend limit before the first developer signs in, because the Claude Code workspace exists from that moment. On a cloud provider, set the cloud budget control, and add a Claude apps gateway when per-person caps are needed, since the cloud bill attributes everything to one shared credential. In every case, tell developers which message they will see and that /usage-credits or a wait for the reset is the right response, because the page on what each limit message means shows how easily a spend limit is confused with a rate limit or a context warning.

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, and every message, screen name and version number on this page is from Anthropic's documentation read on 4 October 2026. If you want help choosing the right caps for a team, contact Reveneau.