Subscription seats or pay per token: the three ways a company gets Claude Code

A company gets Claude Code, Anthropic's coding tool, in one of three ways, and the way it chooses decides how usage is measured, where spend appears and where it can be capped. Anthropic's documentation on managing costs, read on 4 October 2026, names the three: a Claude for Teams or Enterprise plan, the Claude Console, or a cloud provider [1]. This page explains each in plain words, gives Anthropic's list prices per token, and says which setup Reveneau recommends for which situation. It is part of the guide to Claude Code costs for teams.

Two terms come up on every line. A token is a piece of text the model reads or writes; Anthropic's pricing page estimates one token at 4 characters or 0.75 of an English word [2]. An API is the interface that programs such as Claude Code use to send requests to a model and receive answers. Anthropic's documentation says Claude Code "charges by API token consumption" [1].

Setup one: a Teams or Enterprise plan with seats

On a Claude for Teams or Enterprise plan, each member has a seat, which is a paid place on the plan. Anthropic's documentation, read on 4 October 2026, says each member's Claude Code usage "draws from a per-seat allowance that resets on a rolling five-hour window and a weekly window" [1]. The allowance is the amount of usage included in the seat's price. Two further facts shape a budget. The allowance is shared with Claude chat and Cowork, two other Anthropic products, and its size depends on the seat tier, Standard or Premium [1]. And Anthropic says usage inside the allowance "isn't metered in dollars" [1].

The controls for this setup are in the claude.ai admin console, which Anthropic distinguishes from the Claude Console [1]. The two are different websites with different jobs: the claude.ai admin console manages a subscription organisation, and the Claude Console manages an API account.

When a member uses up the allowance, the request stops with a message such as "You've hit your session limit" or "You've hit your weekly limit", and the message shows when the window resets [1]. To let members continue past the allowance, an administrator turns on usage credits and sets spend limits [1]. Usage credits are paid usage, measured in dollars, that a member draws on after the allowance runs out; Anthropic's documentation describes them as a way to "keep working past your plan's usage limit" [1]. Once usage credits are on, the spend report in organisation analytics shows estimated usage-credit spend per user and per model [1]. The page on spend limits for an organisation, a group and a person covers the limits.

One cost mechanism changes when a member moves onto usage credits. Claude Code reuses already-processed request text through a prompt cache, a store the model service keeps for a limited time. Anthropic's documentation says the cache lifetime "is an hour on a subscription and drops to five minutes once you're drawing on usage credits" [1]. A developer who pauses for 20 minutes while on usage credits therefore reprocesses the full conversation on the next message. The sister guide on cache lifetime covers the setting that keeps the one-hour lifetime.

Some requests bill to usage credits even while the allowance has room. Anthropic's model documentation says that, depending on the plan and seat tier, usage of the Fable models can bill to usage credits instead of the plan's included limits, and the model picker then shows "Requires usage credits" on that row [3]. The same page says the 1 million token context window for Claude Sonnet 4.6 requires usage credits on every subscription plan, while the same window for Claude Opus 4.6 is included on Max, Team and Enterprise plans, including both Team Standard and Team Premium seats [3].

Subscription plan prices are published at claude.com/pricing, which Anthropic's documentation refers to [1]. That page was outside the documents read for this guide, so no seat price appears here.

Setup two: the Claude Console, billed per token

The Claude Console is Anthropic's web console for API accounts. On it, Anthropic's documentation says, usage "is billed per token to your organization" [1]. Every token is metered, and the Console usage page is, in Anthropic's words, the place for "authoritative billing" [1].

The Console organises spend through workspaces, which are named compartments inside an API account that each have their own spend limit and usage reporting. 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 in the organisation, and that no API keys can be created for it, because it exists only for Claude Code sign-in and usage [1]. An API key is a secret string that lets a program use the account. Spend limits are set on that workspace [1].

Prices are per million tokens and differ by model and by token type. The table gives Anthropic's list prices from its pricing page, read on 4 October 2026, for the models most likely to appear in a Claude Code budget [2]. Input tokens are text the model reads. Output tokens are text the model writes, including its reasoning. A cache write stores request text for reuse, and a cache read reuses it.

Model	Input, per million tokens	Five-minute cache write	Cache read	Output, per million tokens
Claude Fable 5.1	$10	$12.50	$0.25	$50
Claude Opus 5.5	$4	$5	$0.20	$20
Claude Sonnet 5.5	$2	$2.50	$0.20	$10
Claude Sonnet 4.6	$3	$3.75	$0.30	$15
Claude Haiku 4.5	$1	$1.25	$0.10	$5

Two details from the same page matter for a Console budget. For Claude 4.6 and later models, choosing US-only inference through the inference_geo parameter adds a 1.1 multiplier to every token price, and Claude Code applies that multiplier in its own cost figures from version 2.1.239 [2][1]. And volume discounts are, in Anthropic's words, "negotiated on a case-by-case basis", and enterprise customers can contact sales for custom pricing [2]. If your company has a contract rate, the page on where to see Claude Code spend explains the setting that makes Claude Code report at that rate.

Billing, by Anthropic's pricing page, is based on actual monthly usage, in US dollars, with credit card and invoicing options [2]. New users receive a small amount of free credits to test the API [2].

Setup three: a cloud provider, billed to the cloud account

The third way is through a cloud provider: Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. These are services run by Amazon, Google and Microsoft that offer Claude models inside the customer's existing cloud account. Anthropic's documentation says that on these providers Claude Code "is billed per token to your cloud account, and spend controls live in your cloud provider's billing console" [1].

Two consequences follow. First, the prices are set outside Anthropic's list. Anthropic's pricing page describes Amazon Bedrock and Google Cloud as partner-operated platforms where the cloud provider invoices you, says the two have independent regional pricing, and refers readers to each provider's own pricing page [2]. Microsoft Foundry, by the same page, bills through the Azure Marketplace in Claude Consumption Units: Anthropic rates token usage in US dollars at its standard per-model rates, applies any negotiated discount, and converts the result to units at $0.01 per unit, so the Azure bill shows one line [2]. Neither cloud provider's pricing page was read for this guide, so no Bedrock or Google Cloud price appears here. Second, visibility is different. Anthropic's documentation says Claude Code "does not send metrics from your cloud back to Anthropic", so the analytics dashboards (web pages of figures and charts) and the Claude Code Analytics API do not cover this usage [1]. Per-user reporting on a cloud provider comes instead from OpenTelemetry export, a standard way for each developer's machine to send usage measurements to a monitoring system the company runs, or from a gateway, a server the company runs between Claude Code and the model provider; the page on per-developer usage reporting explains both.

The model behind a name also differs by provider. Anthropic's model documentation, read on 4 October 2026, says the opus alias resolves to Opus 5.5 on the Anthropic API, on Amazon Bedrock and on Google Cloud's Agent Platform, and to Opus 4.6 on Microsoft Foundry; the sonnet alias resolves to Sonnet 5.5 on the Anthropic API and to Sonnet 4.5 on all three cloud providers [3]. A budget built on the Anthropic API prices for Sonnet 5.5 will be for a different model on Bedrock.

Mixed sign-in and gateways

Companies often end up with more than one setup at once. Anthropic's rule for that case is one sentence: "If your organization mixes sign-in methods, each developer is metered according to the one they authenticated with" [1]. A developer signed in with a Teams seat draws on the allowance; a developer using a Console API key is billed per token. The administrator therefore has to know, per developer, which sign-in is active.

A gateway changes the answer again. A gateway is a server your company runs between Claude Code and the model provider, so that developers authenticate to the gateway and the gateway forwards requests with one company credential. Anthropic's gateway documentation says that when developers connect through a gateway with a gateway credential, "usage is billed to your organization's provider account at API rates, and their claude.ai subscriptions aren't used or charged" [4]. The exception is a machine where only the gateway address is set and no gateway credential: requests still route through the gateway, but the saved subscription login stays active and its usage limits and billing apply [4].

The three setups compared

	Teams or Enterprise plan	Claude Console	Cloud provider
How usage is metered	Per-seat allowance on a five-hour and a weekly window; dollars only for usage credits [1]	Per token, billed to the organisation [1]	Per token, billed to the cloud account at the provider's prices [1]
Where to see spend	Spend report in organisation analytics, updated daily, once usage credits are on [1]	Console usage page for billing; Console dashboard for daily costs and per-member spend [1]	The cloud provider's billing console [1]
Where to cap spend	Usage credits off, or spend limits at organisation, group or member level in the claude.ai admin console [1]	Spend limit on the Claude Code workspace [1]	The cloud provider's budget controls [1]
Per-user numbers	Spend report CSV (a spreadsheet file); Enterprise Analytics API on Enterprise [1]	Console dashboard and Claude Code Analytics API [1]	OpenTelemetry or a gateway [1]
Anthropic's dashboards cover it	Yes [1]	Yes [1]	No, because no metrics return to Anthropic [1]

Which setup to pick

Reveneau's recommendation, from the facts above. Pick a Teams or Enterprise plan when the company wants a known monthly figure and a built-in upper limit: the allowance stops usage by default, and dollars only appear if an administrator turns usage credits on. Pick the Claude Console when the company already has an API account, wants every token metered in dollars, and wants the per-user reporting of the Console dashboard and its Analytics API. Pick a cloud provider when procurement or data rules require spend to be billed to an existing Amazon, Google or Microsoft account, and accept that per-user reporting then has to be built with OpenTelemetry or a gateway. In every case, run the pilot described on the page on what Claude Code costs per developer on the setup you intend to keep, because the numbers come from different screens in each.

Reveneau is an AI software development consultancy. All of its code is written by AI, so token use is a running cost of every Reveneau build, and Reveneau is independent of Anthropic. Every price on this page is Anthropic's list price as published on 4 October 2026. If you want help choosing a setup for your team, contact Reveneau.