Claude Code Pricing with Subscription Plans and What Claude Code Costs

Claude Code costs $20 a month on the cheapest plan that includes it, Claude Pro, or $17 a month if you pay $200 for a year up front. Heavier individual users pay $100 for Max 5x or $200 for Max 20x. Companies pay $20 to $25 per Team Standard seat, $100 to $125 per Team Premium seat, or $20 per Enterprise seat plus every token at API rates. Prices exclude sales tax.

What those prices buy is not a number of prompts. Anthropic's pricing FAQ says every plan has usage limits that reset on a rolling five-hour session window, that paid plans add weekly limits on top, and that "your activity across Claude on web, desktop, mobile, and Claude Code all draws from the same pool." A developer who also uses Claude for research and writing is spending one allowance on both jobs. Anthropic gives no fixed message count because consumption depends on context size, the model and the features used.

The cost unit that does exist is the multiple. Pro is at least 5 times Free per session, Max is 5 or 20 times Pro, a Team Standard seat is 1.25 times Pro per Anthropic's Team plan help article, and a Premium seat is 5 times Standard. Divide each price by its multiple and you get the price of one Pro-sized session allowance on every plan:

The pattern is exact: Anthropic charges $20 per Pro-sized allowance on every monthly plan and $16 to $16.67 on annual ones, with one exception. Max 20x charges $10, half the rate of everything else, because it doubles the price of Max 5x and quadruples the usage. For one developer who lives in Claude Code all day, Max 20x is the cheapest capacity Anthropic sells. It is an individual plan, though, and it cannot be bought as a Team seat. The table covers per-session capacity only; Anthropic does not publish weekly limits as multiples, so treat the weekly cap as the unknown in every row.

One change since September makes Max cheaper than this table suggests. Anthropic's plan comparison now lists monthly API credits on Max: $100 a month on Max 5x and $200 on Max 20x, for building on the Claude Platform. The Team plan carries up to $500 a month in pooled API credits; Pro and Enterprise list none. If you would otherwise pay for API tokens for scripts, CI jobs or a small internal agent, those credits offset part of the Max price, and on Max 20x the listed credit equals the plan price. Anthropic marks these credits as new, so check the terms on your billing page before you count on them.

Buy Pro if you use Claude Code in focused bursts and can live with an occasional wait for the five-hour window to reset. Buy Max 5x at $100 if you hit the Pro session limit more than a couple of times a week, and Max 20x at $200 if you run long agent sessions most of the day or run several sessions in parallel. Anthropic's FAQ says both Max options are billed monthly, and upgrades take effect immediately with unused time credited.

There is one difference between the plans that the price table does not show: which models you get inside the allowance. Anthropic's help article on Fable models says Fable 5 and Fable 5.1 are available on all paid plans, but the terms differ by seat:

For a developer who wants Fable for hard refactors, that is a real price step. Fable 5.1 is two and a half times the list price of Opus 5.5, so a few heavy Fable sessions on Pro credits can cost more than the $80 gap between Pro and Max 5x. If Fable is part of your workflow, Max or a Premium seat is the plan, not Pro plus credits.

A subscription is cheaper than the API for anyone who uses Claude Code on more than a handful of days a month and stays inside the limits. The API is cheaper for occasional use, and it is the only path with no ceiling, which is why Anthropic suggests it for heavy sessions. On the API, Claude Code bills every token at list price through a Claude Console key, or through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry, where it lands on your cloud bill.

Anthropic's Claude Code cost documentation gives the benchmark that makes the comparison possible: "across enterprise deployments, the average cost is around $13 per developer per active day and $150-250 per developer per month, with costs remaining below $30 per active day for 90% of users." Set that against the plan prices and the break-even points fall out:

A developer at Anthropic's average who works 20 days a month costs about $260 on the API. The same person on Max 20x pays $200 and, on the capacity math above, has more headroom per dollar than on any other plan. The API only wins below roughly eight active days a month on Max 5x usage, or when the weekly limit, not the price, is what stops you.

Two API details change the numbers. Anthropic bills responses served under data residency at 1.1 times list, and Claude Code applies that multiplier in its own cost display from version 2.1.239. And prompt caching lasts five minutes by default on an API key, against an hour on a subscription, so a developer who steps away for a meeting pays to rebuild the cache more often on the API.

A mid-sized agent task in Claude Code costs roughly $1.50 on Sonnet 5 and $2.55 on Opus 5.5 at list API prices, using the token mix below. The mix is our own illustrative assumption, not an Anthropic figure: an agent working through a feature in an existing repository that reads 2 million tokens from cache over its turns, writes 150,000 new tokens to cache, sends 50,000 uncached input tokens and returns 60,000 output tokens, thinking included.

Three things are worth noticing. First, Opus 5.5 lists at twice Sonnet 5, but the task costs only 1.73 times as much, because both models charge the same $0.20 per million for cache reads, and in a long Claude Code session cache reads are most of the tokens. Second, output is the biggest line on every model, and extended thinking is billed as output; Anthropic notes you cannot turn thinking off on Opus 5.5 or Fable, so lowering the effort level is the lever there. Third, about 20 of these Sonnet tasks in a day comes to roughly $30, the level Anthropic says 90% of users stay under.

Anthropic lists the habits that cut the bill most: clear the session between unrelated tasks, default to Sonnet and reserve Opus for architecture, keep MCP servers you are not using switched off, and hand verbose work such as test runs to subagents on a smaller model. Agent teams are the exception to budget for: Anthropic says they use about 7 times more tokens than a standard session when teammates run in plan mode.

For a company, Claude Code comes with Team seats or Enterprise seats, and the two bill in opposite ways. A Team seat includes a usage allowance with a ceiling; members who pass it stop, or continue on usage credits if an owner has turned them on. An Enterprise seat is $20 of access, and every token behind it is billed at API rates with no per-seat limit, so the ceiling is whatever spend limit an admin sets. Team runs from 2 to 150 seats. Enterprise is annual only, and Anthropic sells it self-serve or through sales.

Here is the monthly bill for a 10-developer team on each shape, before usage credits:

The Enterprise break-even is simple to work out. A Premium Team seat costs $100 a month on annual billing, and an Enterprise seat costs $20 plus usage, so Enterprise is cheaper per developer only while that developer's API usage stays under $80 a month. At Anthropic's $13 average that is about six active days. A developer who uses Claude Code most working days costs $170 to $270 a month on Enterprise, against $100 on a Premium seat that covers them until the ceiling. Companies usually move to Enterprise for SCIM, audit logs, the compliance API, custom data retention and network controls rather than for price, and they should budget it as a metered line.

Controls differ too. On Team and Enterprise, Claude Code usage shows up in the spend report in org analytics with a per-user CSV, and admins can cap usage credits at the organization, group or member level. Enterprise adds the Enterprise Analytics API for per-user cost pulls. Our Claude Enterprise pricing breakdown covers seat minimums and the rest of the Enterprise contract, and Claude Team vs Enterprise pricing runs the seat math at larger headcounts.

When you hit a Claude Code usage limit you can wait for the window to reset, move to a larger plan, or turn on usage credits and keep working at standard API rates. Claude Code tells you which limit you hit: a session limit resets on the rolling five-hour window, a weekly limit at a fixed time each week, and switching models does not get around either, because both apply across all models.

One detail from Anthropic's docs is easy to miss and costs real money: on a subscription the prompt cache lasts an hour, but once you are drawing on usage credits it drops to five minutes unless you choose the TTL yourself. A developer on credits who pauses for ten minutes pays to reprocess the whole conversation on the next message. If a team runs on credits for more than a few days a month, the higher plan is almost always the cheaper fix. Inside Claude Code, /usage shows plan usage, credit spend and which skills, subagents and MCP servers are consuming it, and /usage-credits opens the credit settings.

Claude Code is the cheapest way to run Claude models for coding if you already pay for a Claude plan, because Anthropic adds no markup on its own models. Cursor charges its Teams and Enterprise customers a $0.25 per million token rate on top of third-party models, which adds about $0.57 to the Sonnet 5 task above, a 38% increase. GitHub Copilot bills model use in $0.01 AI credits at the model's rate, with $19 of credits included in a Business seat.

The seat prices are close; the usage terms decide the bill. We broke down each vendor in Cursor pricing, GitHub Copilot Enterprise pricing and the head to head in Claude Code vs Cursor pricing.

Before you size seats, look at what the team actually runs through Claude Code. A good share of it is repeated work that does not need a developer driving a frontier model: reviewing every pull request against your conventions, turning bug reports into labeled and routed tickets, keeping API docs in step with the code. A ready-made agent that does one of those jobs runs on a flat monthly price with no tokens and no weekly limit to watch. The coding agents at the top of this page, Code Reviewer, Bug Triager and Docs Writer, take that work off your Claude Code allowance so it goes to the tasks that need it. For the wider comparison of agent vendors and how they meter, see AI agent platforms compared.