Claude Code pricing: Pro, Max, API costs and usage limits

Claude Code pricing starts at $20/month with Pro, or $200 paid annually. Max costs $100/month for 5x Pro’s per-session allowance or $200/month for 20x. Team has Standard and Premium seats; current Enterprise plans charge a seat fee plus metered usage. You can also pay through the API. Anthropic’s pricing page lists the current plans, checked October 7, 2026.

The decision is how much work you can finish before a limit interrupts you, and what happens to the bill when you continue. Subscription allowances, API tokens, and the context window measure different things. Permute publishes this breakdown and sells connected business data with supported custom app delivery. We use coding agents as part of engineering work; buying more agent capacity still leaves someone responsible for the finished application.

Claude Code plans and monthly prices

Pro’s annual price works out to $16.67/month and saves $40 against twelve $20 payments. Both Team seat types include Claude Code, and the current minimum is two members. Free Claude does not include Claude Code. Prices above are in US dollars; confirm taxes, currency, and the billing cycle at checkout.

Is Max 5x or Max 20x worth the upgrade?

Anthropic defines Max’s multipliers against Pro’s per-session allowance. Moving from monthly Pro to Max 20x raises the price from $20 to $200, a 10x increase, for a stated 20x session allowance. That arithmetic favors Max on allowance per dollar. It does not establish twenty times as many completed features, a fixed token balance, or a twentyfold weekly allowance.

Record a week of representative work before upgrading. Note which limit stopped you, how much time the interruption cost, and whether the output passed review. If Pro repeatedly runs out during useful coding sessions, a larger session allowance can help. If the weekly ceiling ends your work on Thursday, assess that constraint separately rather than relying on the plan’s 5x or 20x label.

A larger plan also gives more room for expensive mistakes. An agent can spend a long session following a wrong assumption or repairing a change you will discard. Compare accepted patches and review time alongside usage; tokens consumed alone do not show what the subscription bought.

Five-hour limits and weekly resets are separate

Pro’s limit documentation describes rolling five-hour session windows and an all-model weekly ceiling. The weekly reset occurs at a fixed time assigned to your account, regardless of when you begin work or subscribe. Max also has both limits. Check the displayed reset time rather than assuming the allowance returns at midnight or seven days after your last request.

For example, a session that refreshes at 2 p.m. does not make coding available then if your all-model weekly allowance is already exhausted. Opening another terminal or starting a new conversation does not create a separate account allowance. Changing models cannot bypass an all-model ceiling; a model-specific ceiling is a different restriction, so read the message identifying what blocked the request.

Claude chats and Claude Code consume the same account limits. A morning of document analysis can therefore reduce the capacity left for afternoon development. Anthropic’s usage explanation also distinguishes context length from usage: the context window limits what a conversation can hold, rather than promising that many tokens every month.

Read /usage without confusing it with your bill

Claude Code’s cost documentation explains the distinction. /usage shows plan-limit status alongside token use and estimated cost. The session dollar estimate uses model pricing; it is not an extra subscription invoice. The day and week attribution views use local machine history, so they exclude activity on other machines and in Claude chats. Use billing records to reconcile paid consumption.

/clear starts a fresh session and resets its displayed cost, without restoring your plan allowance. /usage-credits opens paid-usage controls when you are signed in through a subscription. Before a long run, check both the current session bar and the weekly bar, then confirm whether paid continuation is enabled.

API access and usage credits create additional charges

Usage credits let Pro and Max subscribers continue beyond included limits at standard API rates. You enable them, add funds, and set a monthly spending limit. Auto-reload can replenish the balance when it falls below your chosen threshold. These charges sit on top of the subscription, and paying for overflow does not move the normal reset time.

Check how the terminal authenticated. Anthropic’s subscription setup instructions warn that an ANTHROPIC_API_KEY environment variable takes precedence over subscription login and bills API usage. Use /login to select the intended account path, and inspect whether a key is configured without printing its value. A Max subscription does not cover requests billed to a separate Console account.

Current API pricing charges Sonnet 5.5 $2 per million input tokens and $10 per million output tokens; Opus 5.5 costs $4 and $20 respectively. A hypothetical request with 100,000 uncached input tokens and 10,000 output tokens costs $0.30 on Sonnet or $0.60 on Opus. That calculation excludes cache pricing, tools, and other charges. An agent completing one task may make many such requests.

Parallel chats spend the same allowance faster

Three active Claude Code sessions under one login share the account’s capacity. They do not each get a new five-hour allowance. Agent teams add another consideration: each teammate has its own context window, and token use grows with the number of active agents and how long they run. Parallelism can shorten elapsed time while increasing consumption.

Start with independent tasks that can be reviewed separately, such as a frontend change and a test update. Give each agent a bounded file scope and completion check. If several agents need to edit the same function, coordination and rework can erase the time saved. There is no reliable universal number of parallel chats or coding tasks per day for a subscription.

Model choice, effort level, conversation length, and tool activity affect consumption, according to Anthropic’s usage recommendations. Keep relevant repository instructions, summarize completed work when changing tasks, and avoid carrying unrelated logs into every request. Measure the effect on your workload before treating a context trick or a cheaper model as a promised capacity increase.

Budget Team seats separately from Enterprise consumption

Team’s current plan assigns each member an allowance. Standard provides 1.25x Pro’s per-session usage; Premium provides 6.25x, or five times Standard. Mix seats around measured needs rather than buying Premium for everyone. Three Standard seats and one Premium cost $200/month on monthly billing, or $1,920 annually at the published annual rates.

Current Enterprise billing separates seat access from consumption: the seat fee includes no token allowance. Self-serve organizations share prepaid credits; sales-assisted organizations generally receive usage invoices in arrears. Organization and individual spending limits can stop work even when access remains active. Older Enterprise contracts can have included seat allowances, so check the actual contract before applying today’s model.

Check reset offers and promotions before relying on them

Anthropic sometimes offers a limit reset. The offer specifies whether it restores session or weekly capacity and when it expires. Redeem it on the web or desktop; the restored account allowance also applies to Claude Code. A weekly reset offer does not shift your usual weekly reset time or add paid credits. Treat it as temporary capacity rather than part of the recurring plan.

Read promotion exclusions. As of October 7, 2026, the October 1-15 artifact promotion reduces qualifying Claude and Cowork work’s five-hour usage by 50%, but excludes Claude Code and leaves weekly limits unchanged. A headline about doubled capacity elsewhere in Claude is therefore insufficient evidence for a larger coding allowance.

If another coding vendor has spare quota, switching may keep a task moving without buying overflow. Carry over the goal, current branch, completed checks, and next step (not every line of the old chat). Our Codex vs Claude Code comparison covers current plan differences and how to assess usable capacity across the two. Recheck terms when a model or plan changes.

Permute: business data and supported app delivery

Permute sells connected business data and custom apps with engineering implementation and support. We reconcile records across approved sources, apply agreed definitions, and deliver apps around that data. A job dashboard needs a rule for matching the same project across accounting and operating systems, plus an owner for fixing mismatches. A coding subscription supplies engineering capacity; that data work belongs in the delivery scope.

Claude Code can help a developer build the interface while Permute supplies governed data and supports the application. Our custom software offering covers the delivery path. Budget the agent subscription alongside implementation and maintenance rather than comparing the subscription fee with the price of a finished business app.

Keep coding access separate from deployed app access

Permute’s hosted apps serve externally built, compiled static frontends. Queries and actions check current user permissions and app grants. Builds, source editing, and server runtimes remain outside that hosting scope. ERP writes need a separately supported, approved workflow. Our security page describes the controls to review for confidential business records.

Before launch, assign responsibility for data access tests, deployment, and ongoing changes. A coding agent’s account settings do not define which jobs or properties an employee can see inside the finished app. The enterprise app development article explains the handover work that belongs beside the coding budget.

Choose a plan from measured interruptions and reviewed output. Confirm the reset schedule, authentication path, and overflow cap before starting sustained work. Revisit the budget when the model, workload, or contract changes. Assign an operating owner to the finished app so increased coding capacity leads to software people can rely on.

Questions about Claude Code pricing