Claude Code pricing starts at USD 20 a month on Claude Pro, runs to USD 100 or USD 200 a month on Claude Max, and costs USD 25 per seat a month on a Claude Team plan; the free Claude plan does not include Claude Code at all. Pay per token through the Claude API instead and the Anthropic cost documentation puts a typical developer at around USD 13 per active day. Every figure below was read off the Anthropic pricing page, Help Center and Claude Code docs in August 2026. By the end you will know which tier fits the agent time you actually use, and which one to skip.
Part of: How to Use Claude Code: Complete Beginner Guide (2026)
How to read the numbers on this page
There is no single Claude Code price, because there are three separate billing models and two separate limit clocks. Get those five things straight and the plan decision takes about ten minutes.
- Subscription seats: a flat monthly fee for Claude Pro, Max, Team or Enterprise, with Claude Code usage included until you run into a usage window.
- Usage credits: once a Pro or Max subscription reaches its included limit, you can prepay for extra usage, which Anthropic bills at standard API rates rather than at a subscription rate.
- Pay per token: authenticate Claude Code with a Claude Console API key or a cloud provider and pay for every input and output token. No monthly fee, and no hard stop when you overspend.
- The five-hour clock: every plan has a rolling session window that resets five hours after it starts.
- The weekly clock: paid plans add a weekly limit on top, resetting at a fixed time each week that Anthropic assigns to your account.
- One shared pool: the Claude web app, the Claude desktop app and Claude Code all draw from the same allowance, so a heavy chat day shortens your coding day.
Every figure here was checked against the Anthropic pricing page, Help Center and Claude Code documentation in August 2026. Anthropic revises these plans and limits several times a year and says so openly, so treat each number as a snapshot and re-check before committing to an annual plan.
Is Claude Code free, and is there a free trial?
No. Claude Code is not free, there is no free tier, and the free Claude plan does not include it. The Claude Code documentation is unambiguous: Claude Code requires a Pro, Max, Team, Enterprise or Console account, and the free Claude plan does not include Claude Code access. That plan runs the chat app only.
- The client is free: installing Claude Code costs nothing on macOS, Windows and Linux. What you pay for is the model usage behind it.
- There is no time-limited Claude Code free trial. The nearest thing is the small amount of free credits Anthropic gives new Claude API users, which you can spend through Claude Code with a Console API key.
- The cheapest committed path is Claude Pro at USD 20 a month, or USD 17 a month if you pay USD 200 up front for a year.
- Bring your own cloud account is not a free path either: running Claude Code against Amazon Bedrock, Google Cloud or Microsoft Foundry drops the Anthropic subscription but bills every token to your cloud provider.
Who should not buy a subscription: if you want one evening to find out whether directing an agent suits you, a Console API key and the free starter credits cost less than a month of Pro.
Claude Code plans and prices, tier by tier
Claude Code Pro pricing is the simplest line here. The Claude Code Pro price is USD 20 a month, or USD 17 a month if you pay USD 200 up front for a year. Above that, Max and Team seats buy the same tool with a larger allowance, and the difference between every tier is how much agent time you get before a window closes.
- Claude Pro. Price: USD 20 a month, or USD 17 billed annually. Usage: at least five times the per-session usage of the free service. Watch out for: a weekly cap sits on top of the five-hour window.
- Claude Max 5x. Price: USD 100 a month, monthly billing only. Usage: five times more per session than Pro. Watch out for: no annual discount exists, so the sticker price is the real price.
- Claude Max 20x. Price: USD 200 a month, monthly billing only. Usage: 20 times more per session than Pro. Watch out for: the weekly limit still applies, so 20x per session is not unlimited.
- Claude Team standard seat. Price: USD 25 per seat a month, or USD 20 billed annually. Usage: a per-seat allowance larger than Pro. Watch out for: the allowance is shared with chat, not reserved for coding.
- Claude Team premium seat. Price: USD 125 per seat a month, or USD 100 billed annually. Usage: five times a standard seat. Watch out for: it costs more than Max 5x for one person.
- Claude Enterprise. Price: quoted by Anthropic sales. Usage: included with the single Enterprise seat on new and self-serve plans. Watch out for: usage-based Enterprise plans have no per-seat limit and bill at API rates.
Two details are easy to miss. Both Max prices are for web subscriptions, and Anthropic notes that mobile pricing may vary by app store, so subscribe on the web. And the gap between seat-based and usage-based Enterprise plans matters more than any seat price: seat-based plans put each member on the same five-hour and weekly clocks as an individual, while usage-based plans have no allowance to run out of, only a bill.
Claude Code usage limits: the five-hour window and the weekly cap
The Claude Code limitations that matter are not missing features, they are two clocks running at once. The five-hour session window resets five hours after the session starts, and paid plans add a weekly limit that resets at a fixed time each week assigned to your account. Anthropic tracks that weekly reset separately for Opus and for all other models, which is why heavy Opus use can lock you out of Opus while everything else keeps working.
- Everything counts once: usage across the Claude web app, Claude Desktop and Claude Code draws from a single pool, so an afternoon of chat eats the same window as your agent runs.
- Two kinds of limit message: a session or weekly limit applies across all models, and switching model with /model will not restore access. A model-specific message such as an Opus limit can be worked around by switching to a model outside that family.
- You can see the clocks: Settings and Usage in the Claude web app shows progress bars for both windows, and /usage shows the same bars inside Claude Code.
- You can wait it out: on Claude Code v2.1.234 or later, Claude Code can wait for the reset and continue the interrupted task automatically instead of dropping the work.
- You can buy through it: Pro, Max 5x and Max 20x subscribers can turn on usage credits and keep going at standard API rates, with a daily redemption cap of USD 2000. Team and Enterprise admins enable the same thing and set spend limits per organization, group or member.
- On an API key there is no window at all: nothing stops, and the account is charged for what it uses.
Anthropic does not publish how many hours of agent time a five-hour window buys, and no honest page can tell you: it depends on your model, your context size and how many tool calls each turn makes. What Anthropic does publish is the shape of the limit, and the shape is what you plan around. Start expensive work early in a window, not at the end of one.
What Claude Code costs for three real workloads
Price your own workload twice, once as a subscription and once per token. Anthropic publishes the figures that make this possible: across enterprise deployments the average is around USD 13 per developer per active day and USD 150 to 250 per developer per month, with 90 percent of users below USD 30 per active day. Here is that arithmetic for three situations, at 20 active days a month.
- One founder, a few hours of agent time most working days. Per token: 20 days at USD 13 = USD 260 a month. Subscription: Pro at USD 20, or USD 17 on annual billing. Pick Pro, and move to Max 5x at USD 100 only once the five-hour window starts interrupting real work.
- A two-person team shipping a product. Per token: 2 developers x 20 days x USD 13 = USD 520 a month. Subscription: two Team standard seats at USD 25 = USD 50 a month, or USD 40 billed annually. Pick Team seats, for the central billing and admin controls two personal Pro accounts cannot give you.
- One heavy user running long agent sessions most of the day. Per token at the 90th-percentile ceiling: 20 days x USD 30 = USD 600 a month. Subscription: Max 20x at USD 200 a month. Pick Max 20x, and watch the weekly cap rather than the price.
The same arithmetic at the request level shows where the money goes. One Claude Opus 5 request carrying 50,000 input tokens and producing 15,000 output tokens costs 50,000 at USD 5 per million, which is USD 0.25, plus 15,000 at USD 25 per million, which is USD 0.375: USD 0.625 in total. Serve 40,000 of those input tokens from the prompt cache at USD 0.50 per million and the same request costs USD 0.445. Output tokens dominate and cached input is nearly free, which is exactly why the levers below work.
How to cut your Claude Code bill
Five levers move the number, and the Claude Code cost documentation states the direction of each one. None of them require a plan change. You also do not need a third-party Claude Code usage monitor: the /usage command reports your plan bars, your recent token spend, and which skills, subagents and MCP servers are responsible for it.
- Match the model to the task. Sonnet handles most coding work and costs less than Opus; Haiku is cheaper again for simple subagent jobs. At August 2026 API rates, Claude Sonnet 5 is USD 2 per million input tokens and USD 10 per million output, against Claude Opus 5 at USD 5 and USD 25.
- Clear between unrelated tasks. Claude Code sends the full conversation with every request, so a one-line question in a session you opened this morning still draws usage for the whole morning. The /clear command costs nothing.
- Protect the cache. Prompt caching is automatic and cache reads cost a tenth of the base input price, but the cache lifetime is one hour on a subscription and drops to five minutes once you are drawing on usage credits.
- Turn thinking down when you do not need it. Extended thinking is on by default and thinking tokens are billed as output tokens, the expensive ones. The /effort command lowers it per task.
- Watch what sits in context. Run /context to see what is consuming space, disable MCP servers you are not using, keep CLAUDE.md to the essentials, and push specialized instructions into skills that load only when invoked.
One number worth knowing before you scale up: Anthropic states that agent teams use approximately seven times more tokens than standard sessions when teammates run in plan mode, because each teammate carries its own context window. Background token use, by contrast, is trivial, typically under USD 0.04 per session. Context discipline is the single biggest lever on this bill, and it is what the free courses at Agentic School spend most of their time on.
Subscription or API: which one is cheaper for you
For one person coding daily, the subscription wins on price and loses on predictability. What you give up for the saving is the ability to keep working once a window closes.
- Choose a subscription when usage is steady, you work alone or in a small team, and you can live with a hard stop until the next reset.
- Choose an API key or a cloud provider when the work is bursty, automated or must not stop mid-run, and when you need per-user spend attribution more than a flat fee.
- Choose both when you have automation: a subscription for interactive work, and a separate Console key for scheduled or CI jobs that would otherwise burn your personal window.
The honest downside of the token-billed route is that nothing stops: no window, no warning and no cap unless you set a workspace spend limit yourself. That is the right trade for a scheduled job and the wrong one for somebody still learning the tool.
Claude Code plans compared
One line per option, in the order you should consider them.
- Free Claude plan: USD 0, and no Claude Code at all. Best for chat only. Watch out for: it is not a trial, the tool is simply not included.
- Claude Pro: USD 20 a month, or USD 17 billed annually. Best for one person coding a few hours a day. Watch out for: the weekly cap on top of the five-hour window.
- Claude Max 5x: USD 100 a month. Best for one person whose Pro sessions keep running out. Watch out for: monthly billing only, no annual discount.
- Claude Max 20x: USD 200 a month. Best for long agent sessions most of the working day. Watch out for: 20x per session still sits under a weekly limit.
- Claude Team standard seat: USD 25 a month, or USD 20 billed annually. Best for a small team that wants central billing. Watch out for: the seat allowance is shared with chat.
- Claude Team premium seat: USD 125 a month, or USD 100 billed annually. Best for the heaviest one or two people on a team. Watch out for: it costs more than Max 5x for a single user.
- Claude Console API key: no monthly fee, around USD 13 per active day on average. Best for automation and bursty work. Watch out for: no hard stop, so set a workspace spend limit.
Which plan should you pick
Two things decide this, and price is not one of them: how many uninterrupted hours of agent time you need inside a five-hour block, and whether a hard stop is merely annoying or actually costs you money.
If you are one person starting out, buy Claude Pro at USD 20 a month and pay monthly for the first few months, so you see your own rhythm before locking in the USD 200 annual price. If you are one person hitting the session limit more than once a week, go to Max 5x at USD 100 rather than buying usage credits repeatedly, because credits are billed at standard API rates and stop being cheap quickly. If you are a two or three person team, buy Team standard seats at USD 25 each and add a single premium seat at USD 125 only for whoever runs agents all day. If you are running scheduled jobs or CI, put those on a separate Console API key so they never consume the window you need for interactive work.
And skip Max 20x until Max 5x has actually failed you. USD 200 a month buys 20 times the per-session usage of Pro, but it does not remove the weekly cap, and most people who think they need it are paying to work around a context problem they could fix for free. Getting more out of the plan you already pay for is what we teach across the free courses at Agentic School.