How much does Claude Code cost?

How much does Claude Code cost?

Claude Code costs $20 a month on the Claude Pro plan, $100 a month on Max, $200 a month on Ultra, or a per token rate if you pay with API credits from the Claude Console. There is no free version. A subscription buys a usage allowance that Claude Code shares with Claude chat. API credits buy tokens, billed at a published rate per million tokens, subject to your organization's monthly spend cap and any lower limit you set. Every figure on this page is as of 31 October 2026.

Which path is cheaper depends on how many hours a day you run it, and on whether you need spending reported per user. The numbers for both paths are below, along with a worked estimate for one developer working alone.

Is Claude Code free?

No. Claude Code has no free tier. Anthropic's setup documentation says it directly: Claude Code requires a Pro, Max, Ultra, Team, Enterprise, or Console account, and the free Claude.ai plan does not include Claude Code access. The one other route is a cloud provider such as Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry, which bills per token to your cloud account. You can install the command line tool without paying anything, and claude --version will print a version number, but a free Claude account does not include access to coding sessions. If you already pay for Pro and want to know what that covers before you install anything, what a Pro subscription gets you in Claude Code answers it plan by plan.

Two things get mistaken for a free tier. The first is the free Claude.ai plan, which covers a good deal more than chat: chat on the web, desktop and mobile, web search, creating files and running code, memory across conversations, connecting your apps and tools, Artifacts, and Claude Design, Slides and Docs. Claude Code is not on that list, and Anthropic's pricing page marks it as not included on Free. What the free Claude plan actually includes walks through where that boundary sits. The second is the small amount of free credit Anthropic gives new Claude Console users for testing the API. That is a starter credit, not a monthly allowance, and when it runs out the requests stop.

The cheapest real entry point is Pro at $20 a month, or $17 a month if you pay for a year up front.

Subscription prices as of October 2026

[
  {
    "plan": "Free",
    "price_usd": 0
  },
  {
    "plan": "Pro",
    "price_usd": 20
  },
  {
    "plan": "Pro, billed annually",
    "price_usd": 17
  },
  {
    "plan": "Max",
    "price_usd": 100
  },
  {
    "plan": "Ultra",
    "price_usd": 200
  },
  {
    "plan": "Team, standard seat",
    "price_usd": 25
  },
  {
    "plan": "Team, premium seat",
    "price_usd": 125
  }
]

Pro at $20 a month is the entry plan that includes Claude Code. Max and Ultra are separate plans, both billed monthly only. Max at $100 a month gives five times the usage of Pro every five hours and every week, and Ultra at $200 gives ten times the weekly usage. Ultra replaces Max 20x, and existing subscribers move automatically with no price change. These are web prices; mobile pricing may vary. Team is billed per person, $25 a month for a standard seat and $125 for a premium seat, both cheaper on annual billing. Enterprise is listed at $20 a seat a month, billed annually, plus usage charged at API rates. If you are buying for a group rather than for yourself, what each Team and Enterprise seat actually includes sets those seat prices against the cost of paying per token.

What you buy on a subscription is an allowance, not a token budget you can inspect. Pro and Max have a rolling five hour usage window and a weekly limit on top. Ultra has a weekly limit, with no five hour limit for verified accounts; accounts not yet verified have a temporary limit. On Max and Ultra, Fable also has a separate weekly limit of its own, and that limit sits inside the overall weekly limit rather than beside it. Anthropic's pricing page lets you spend up to 50% of the weekly limit on Fable at no extra cost, Fable uses up allowance faster than other models, and because Fable draws from the same overall limit as everything else you run, you can never use more than the weekly limit in total. On Pro, Fable runs on usage credits from the start. None of these limits has to stop you: with usage credits turned on, once you use up an applicable session limit, the weekly limit or the Fable limit on Max or Ultra, your requests switch to pay-as-you-go pricing on usage credits and keep working until you reach your monthly spend limit, if you set one, or your balance runs out with auto-reload off. Without credits, you wait for the reset. The plan allowance is shared with Claude chat and Claude Cowork, so a morning spent in chat reduces what is left for coding in the afternoon. The full Claude tier chooser compares the plans against each other in detail.

What API credits cost per token

If you sign Claude Code in with a Claude Console account instead, there is no monthly fee. You buy credits and the tool spends them per token. MTok below means one million tokens.

[
  {
    "label": "Opus 5.5",
    "input_per_mtok_usd": 4,
    "output_per_mtok_usd": 20,
    "cache_read_per_mtok_usd": "0.20"
  },
  {
    "label": "Opus 5",
    "input_per_mtok_usd": 5,
    "output_per_mtok_usd": 25,
    "cache_read_per_mtok_usd": "0.50"
  },
  {
    "label": "Sonnet 5.5",
    "input_per_mtok_usd": 2,
    "output_per_mtok_usd": 10,
    "cache_read_per_mtok_usd": "0.10"
  },
  {
    "label": "Sonnet 5",
    "input_per_mtok_usd": 2,
    "output_per_mtok_usd": 10,
    "cache_read_per_mtok_usd": "0.20"
  },
  {
    "label": "Sonnet 4.6",
    "input_per_mtok_usd": 3,
    "output_per_mtok_usd": 15,
    "cache_read_per_mtok_usd": "0.30"
  },
  {
    "label": "Haiku 4.5",
    "input_per_mtok_usd": 1,
    "output_per_mtok_usd": 5,
    "cache_read_per_mtok_usd": "0.10"
  },
  {
    "label": "Haiku 5.5, prompts up to 100,000 tokens",
    "input_per_mtok_usd": 0.1,
    "output_per_mtok_usd": 0.5,
    "cache_read_per_mtok_usd": "0.01"
  }
]

Opus 5 costs $5 per million input tokens and $25 per million output tokens. The newer Opus 5.5 costs less, at $4 and $20. Sonnet 5 costs $2 and $10, the same rates as Sonnet 5.5. Anthropic first sold that Sonnet 5 price as an introductory rate through 31 August 2026, then made it the standard price and cancelled the planned rise to $3 and $15. Haiku 5.5 is now the cheapest row at $0.10 and $0.50 for prompts up to 100,000 tokens, rising to $0.50 and $2.50 for longer prompts. Haiku 4.5 costs $1 and $5.

Max, Ultra and Team plans now come with monthly Claude API credits, $100 on Max and $200 on Ultra, but Anthropic excludes interactive Claude Code sessions and extra usage after you hit your plan's limits. The credits cover noninteractive claude -p sessions and the Claude Agent SDK when you run them yourself with an API key from the Console organization you link to the plan. When authenticated with your Claude subscription instead, those runs draw from your plan's usage limits and do not use the monthly API credits. Runs started by the Claude Code GitHub Action, an IDE extension or the Claude desktop app count as Claude Code usage and are not covered, even with -p.

The headline rate is not the whole bill. Claude Code re-sends your entire conversation with every request, so most of your input tokens arrive as cache reads, which cost a tenth of the base input price on most models: $0.50 per MTok on Opus 5. On Opus 5.5 and Sonnet 5.5 a cache read costs a twentieth of base input, $0.20 per MTok on Opus 5.5. Writing to the cache costs more than base input, 1.25 times the input rate for the five minute cache and 2 times for the one hour cache. Extended thinking is billed as output tokens, which is the expensive column above, so a model left on a high effort setting costs real money on every turn.

Model choice is the biggest lever you hold. Choosing which Claude model to run changes the same piece of work by a factor of five or more.

Which billing path costs less for your usage shape

A subscription fits interactive work. You sit at the terminal, you code for a few hours, you stop. The bill is fixed, and the ceiling you run into is a usage window rather than an invoice you did not expect.

API credits fit work that is shared across people or has to be reported per user. Automation on its own no longer forces token billing: a GitHub Actions run, including a scheduled one, can authenticate with an OAuth token from claude setup-token and draw on a Pro, Max, Ultra, Team or Enterprise subscription, and claude -p still draws from subscription limits when authenticated with that subscription. That token is tied to the subscription of the person who generated it, though, so Anthropic recommends an API key for a secret shared across repositories, and per user cost reporting comes from the Claude Console dashboard or, on Team and Enterprise, the admin spend report. Both paths can cap spend. Workspace spend limits in the Claude Console stop spending at a number you choose, and on Pro, Max and Ultra the subscription price is fixed and you can put a monthly spending cap on any usage credits past the allowance, or set it to unlimited if you prefer no cap.

The full comparison of API billing against a Claude subscription works out where the crossover sits, so this page will not repeat the argument. The short answer: if one person uses Claude Code on most working days, a subscription wins, and the decision is only close for light or very bursty use. If you are still choosing a tool at all, how Claude Code compares with the other coding agents puts the subscription prices next to each other.

A worked monthly estimate for a solo developer

Take one developer working alone, about three hours a day, 20 active days a month, Sonnet as the default model and Opus for the hard parts. Anthropic publishes three figures across enterprise deployments, checked on 31 October 2026: an average of about $13 per developer per active day, $150 to $250 per developer per month, and costs remaining below $30 per active day for 90% of users. Multiply the two daily figures by 20 active days to get a month.

[
  {
    "label": "Pro subscription",
    "price_usd": 20
  },
  {
    "label": "Max subscription",
    "price_usd": 100
  },
  {
    "label": "Ultra subscription",
    "price_usd": 200
  },
  {
    "label": "API credits, daily average times 20",
    "price_usd": 260
  },
  {
    "label": "API credits, $30 a day times 20",
    "price_usd": 600
  }
]

On token billing that developer pays around $260 a month if every one of the 20 days costs the published daily average. Anthropic's own monthly figure is lower, $150 to $250 per developer, so $260 sits just above that published range rather than inside it. $600 is what 20 days would cost at $30 each, the daily figure that 90% of users stay below. Anthropic does not publish it as a monthly percentile, but a developer whose days regularly run past $30 pays more than that. Pro at $20 sits far below both figures. Ultra at $200 sits below the daily average multiplied out but inside the published monthly range.

Read the gap carefully, because it is not free money. Pro does not hand you $150 to $260 of tokens for $20. It gives you an allowance; how quickly you reach the five hour limit depends on the model, conversation length, files and effort level. The honest reading of this table is that Max at $100 a month is the plan most likely to fit this usage shape, and it still costs less than the bottom of the published monthly range for token billing, but usage limits still apply, so a heavy week can stop you mid task on it too. Max and Ultra run the same models, but Ultra doubles the weekly allowance and monthly API credits and removes the five hour limit for verified accounts, so choosing between Max and Ultra comes down to how often you hit session or weekly limits. Every figure here is only half the sum, because what a plan costs matters less than what it saves you, and whether that monthly price pays for itself at your own hourly rate is the calculation worth running before you pick a tier.

How to see what you are actually spending

Run /usage inside a Claude Code session. On a subscription it shows plan usage bars plus a breakdown of which skills, subagents, plugins and MCP servers your recent usage went to, and it flags any behaviour responsible for 10% or more of that usage. Press d or w to switch between the last 24 hours and the last 7 days.

On API billing, the Session block at the top of /usage prints a dollar figure:

Total cost:            $0.55
Total duration (API):  6m 20s
Total duration (wall): 6h 33m 10s
Total code changes:    0 lines added, 0 lines removed

That figure is computed locally from token counts priced at list rates, unless an administrator has set your contracted rates in managed settings, so by default it does not know about promotional pricing or a negotiated discount, and it can differ from your invoice. The authoritative number is the usage page in the Claude Console. The session totals also reset when /clear starts a new session, so a small number after a clear does not mean a cheap day. Reading Claude Code token usage line by line explains what each counter is counting.

When you reach a plan limit on Pro, Max or Ultra, /usage-credits opens the usage settings where you can turn on usage credits and keep working past the allowance at token rates. On a Team or Enterprise plan, the same command opens the organization's usage settings if you hold billing access, and otherwise asks you to confirm and then sends a request to your organization's admins.

What pushes the bill up

A long session costs more than the work in it suggests, because every request carries the whole conversation. Four habits lower the bill more than anything else:

• Run /clear between unrelated tasks. Stale context is re-sent on every message that follows it, so an afternoon of unrelated work in one session pays for the morning again and again.
• Switch models with /model. Sonnet handles most coding work, and its output rate is well below Opus, which is where thinking tokens land.
• Watch the cache lifetime. Returning to a session after a long break misses the cache, so your full context is processed again and written back to the cache at the cache write rate, which is above the base input price. By default, the main conversation's cache lasts an hour within a subscription's included usage and five minutes on usage credits or an API key; you can set either lifetime yourself.
• Keep agent teams small. Each teammate runs its own context window, so token usage grows roughly in proportion to team size, and a team uses roughly seven times the tokens of a standard session when teammates run in plan mode.

Keeping the Claude Code context window small is the same skill as keeping the bill small. How much the agent writes matters just as much, because a model that rewrites a whole file when three lines would do bills every one of those lines at the output rate, which is the case for a skill that holds the agent to the smallest change that works. If you run agents unattended, capping what an AI agent can spend on a VPS matters more still, because nobody is watching the session while it burns tokens.

Does running it on a VPS cost extra?

Yes, and it arrives as a separate bill. Running Claude Code on a server instead of a laptop costs you the server. Claude Code needs 4 GB of RAM or more and a supported operating system such as Ubuntu 20.04 or later, so a small VPS is enough. Install it the same way you would locally:

curl -fsSL https://claude.ai/install.sh | bash
claude --version

A working install prints a version string such as 2.1.211 (Claude Code). If the command reports command not found, the installer put the binary in ~/.local/bin, which your shell has not picked up yet: open a new shell and try again. Then run claude in your project directory and log in through the browser prompt, using the paid account you already have. Your plan follows the account, not the machine, so a VPS session draws on the same allowance as your laptop.

The reason to move to a server is not cost. It is that a server session survives your laptop closing. Running Claude Code on a VPS inside tmux covers that setup, and the permissions to set before you leave an agent running covers what to lock down first. Budget the server on top of your plan, and what a VPS actually costs per month gives the current range.

FAQ

Is Claude Code free to use?

No. Claude Code requires a paid Claude plan (Pro, Max, Ultra, Team or Enterprise), a Claude Console account with API credits, or access through a supported cloud provider such as Amazon Bedrock, which bills per token. Anthropic's setup documentation states that the free Claude.ai plan does not include Claude Code access, even though the install itself works. New Console users receive a small amount of free credit for testing the API, which is not a recurring free tier.

Does a Claude Pro subscription include Claude Code?

Yes. Pro costs $20 a month, or $17 a month billed annually, and it includes Claude Code alongside Claude chat. The usage allowance is shared between them and resets on a rolling five hour window. How quickly you reach that window depends on your work, model and effort level. Max at $100 gives five times Pro usage every five hours and every week; Ultra at $200 gives ten times Pro usage every week and has no five hour limit for verified accounts.

Is a subscription or API billing cheaper for daily coding?

For one person coding on most working days, a subscription is almost always cheaper. Anthropic's published average across enterprise deployments is about $13 per developer per active day, which is roughly $260 across 20 active days, and $150 to $250 per developer per month, against $100 a month for Max. API billing wins when you need per user cost reporting or a credential for shared automation that is not tied to one person's subscription, neither of which a personal subscription provides. A spend cap is not the deciding factor, since Pro, Max and Ultra let you set a monthly spending cap on usage credits.

How do I check what Claude Code is costing me?

Run /usage in a session. On a subscription it shows how much of your plan allowance you have used and which parts of the session consumed it. On API billing the Session block shows a dollar total computed locally at list prices, which can differ from your actual invoice, so treat the usage page in the Claude Console as the real number. The session total resets when you run /clear, so check it before clearing.

What happens when I hit my Claude Code usage limit?

Claude Code tells you which window you reached and when it resets. Switching models with /model does not restore access after a session or overall weekly limit, because those windows are shared across models; switching away from Fable lets you keep working within any remaining overall allowance after reaching its separate limit. Verified Ultra accounts have no five hour session limit, but their overall weekly limit and separate weekly Fable limit still apply. Run /usage-credits to turn on usage credits and continue past the allowance at token rates. Without credits, you wait for the window to reset.