A three-tier translucent glass pyramid of stacked slabs, each tier slightly larger than the one above, with a single thin glowing lime green line rising through the center, standing beside a small glass gauge column on a clean reflective white floor, rendered as a semi-realistic 3D architectural concept illustration

Claude Code Pricing: Plans, Limits and Real Costs

Claude Code has no price of its own. It runs on a Claude plan or an API key, so Claude Code pricing is really a question of how much work you ask of it. In tests on real bug reports from an open-source library, a fix cost $0.185 to $0.264 on Opus 5.5 as Claude Code priced it on a Max plan.

A disclosure first: Claude Code is made by Anthropic. This guide ran no tests of its own. It uses Anthropic’s prices as read by independent trackers, plus one outside study of measured task costs. It also reflects a change from late October, when Anthropic renamed its $200 Max 20x tier to Ultra, so older Claude Code pricing guides are out of date.

Claude Code Pricing at a Glance

Claude Code is included in Pro, Max, Ultra, Team and Enterprise, and it isn’t part of the free plan. Prices below were read from Anthropic’s pages on November 11, 2026.

PlanPriceUsage, as Anthropic describes it
Free$0Claude Code not included
Pro$20 a month ($17 billed annually)The baseline for paid plans
Max$100 a month5x Pro usage every five hours and every week
Ultra$200 a month10x Pro usage every week, no five-hour limit for verified accounts
Team, Standard seat$25 a month ($20 annual)More usage than Pro
Team, Premium seat$125 a month ($100 annual)5x a Standard seat
Enterprise$20 a seat, billed annuallyPlus usage at API rates
API keyPer tokenNo plan limits; every token is billed

Ultra is the renamed Max 20x, and existing subscribers moved over at the same price. Max and Ultra are billed monthly only. Because Claude Code shares each plan with Claude chat, a plan’s price covers both, and heavy chat use leaves less for coding. That’s why general Claude pricing and Claude Code pricing are the same question.

Sources disagree on Team Standard seats. Finout and most trackers say every Team seat includes Claude Code, while one guide published this month says Standard seats don’t. Check Anthropic’s Team page before you buy for a group.

How We Compared Claude Code Pricing

Three kinds of source sit behind this guide. Plan prices and limits come from Anthropic’s pages as read by BenchLM on November 11, with AutoRouter’s November 9 check as a second reading. Per-task costs come from AutoRouter’s two studies, which publish their method and data. Budgeting averages come from Anthropic’s cost docs. We didn’t use a paid account or run a benchmark, so where sources disagree, the section says so. Claude Code’s maker is also the maker of the assistant that drafted this guide, which is a reason to read the numbers against the sources linked here.

What Claude Code Costs Per Task

Here is the number most Claude Code pricing guides skip. The figures below come from AutoRouter’s published studies, not from our own testing, and AutoRouter sells a routing app for Claude Code. It discloses that and publishes every run under an open licence. Each figure is the mean of three fresh sessions on Claude Opus 5.5.

TaskLibraryMax-pricedWith an API key
Write a commit messageitsdangerous$0.082$0.061
Fix a planted failing testitsdangerous$0.078$0.056
Rename a function across 6 filesitsdangerous$0.097$0.072
Write tests for one moduleitsdangerous$0.125$0.098
Refactor helpers into a new moduleitsdangerous$0.168$0.132
Fix a real bug: parameter named helpclick$0.237$0.196
Fix a real bug: version_option ignores package_nameclick$0.264$0.221
Fix a real bug: environment variable hintsclick$0.212$0.169
Fix a real bug: empty usage lineclick$0.185$0.150
Write tests for _textwrap.pyclick$0.363$0.306

“Max-priced” is the list-price estimate Claude Code reports on a Max plan, where no money changes hands. The API-key column reprices the same tokens at the cheaper five-minute cache an API key uses. The first five tasks ran in a library of 1,176 lines, and the click tasks in one of about 12,000 lines.

Model choice moves the numbers more than anything else in the study.

ModelAverage task, Max-pricedAverage task, API key
Claude Opus 5.5$0.110$0.084
Claude Sonnet 5.5$0.049$0.037
Claude Haiku 5.5$0.0044$0.0035

All 45 runs in that comparison passed their checks, but they were small tasks. On Terminal-Bench 4.0, Anthropic reported, per AutoRouter, 39.2 percent for Haiku 5.5 against 70.6 percent for Sonnet 5.5, so a cheap pass on a commit message says little about a hard refactor.

Treat these as a floor for simple jobs, not a quote. The libraries are small, the sessions were fresh, and the same task varied by up to about 1.5 times between runs. A long interactive session re-reads a growing conversation and costs more.

To measure your own work, run a task with claude -p --output-format json. The result includes total_cost_usd and token counts. In an interactive session, /usage shows plan usage bars and a breakdown. On Pro, Max and Ultra, the dollar figures are list-price estimates, not bills.

What Drives the Cost: Model, Cache and Context

Claude Code API pricing is simple. With an API key, Claude Code bills every token at Anthropic’s list prices.

Model, per million tokensInputOutputCache read
Claude Fable 5.2$8$40$0.20
Claude Opus 5.5$4$20$0.20
Claude Sonnet 5.5$2$10$0.10
Claude Haiku 5.5$0.10$0.50$0.01

Unless you change it, every request runs on Opus 5.5. That is the default model on every plan and on the API, according to Claude Code’s model docs as summarized by AutoRouter in October. A one-line commit message therefore runs on the most expensive of the everyday models unless you pick another.

The surprise is where the money goes. Writing to the prompt cache was the largest cost on every model in the study: 63 percent on Opus 5.5, 66 percent on Sonnet 5.5 and 51 percent on Haiku 5.5. A fresh session writes Claude Code’s system prompt, tool definitions and the files it reads into the cache, between 6,500 and 16,300 tokens per run. That’s a fixed setup cost each time you start a session.

Cache lifetime also differs by how you sign in. A subscription requests a one-hour cache, with writes at twice the input price. An API key uses a five-minute cache, with writes at 1.25 times. That is why the API-key column above is 19 to 25 percent cheaper, though it only holds because those runs were short.

Claude Code Usage Limits: Five Hours, Weekly and Extra Usage

Anthropic publishes multiples, not token counts. It says how fast you reach a limit depends on the model, the length of your conversations and the size of your files. So you can’t convert a plan into a number of tasks, and nobody outside Anthropic can either. That’s the hardest part of Claude Code pricing to forecast.

What is published is the shape of the limits:

  • A rolling five-hour session limit, except on verified Ultra accounts.
  • A weekly limit on paid plans.
  • One shared pool. Claude chat, Claude Desktop and Claude Code draw from the same limits, so a morning of long chats leaves less for coding.
  • Fable models run on usage credits on Pro, and on Max and Ultra they have their own weekly limit of up to half the overall one.

At a limit you can wait for the reset, upgrade, or turn on usage credits. Usage credits bill at standard API rates, up to a monthly spending limit you set. Credits bought on or after December 7, 2026 expire after 12 months.

Plan or API: Where the Break-Even Sits

A plan price buys a different amount on the API. At list prices, $20 buys 10 million input or 2 million output tokens of Sonnet 5.5, according to BenchLM’s arithmetic. $100 buys 50 million and 10 million, and $200 buys 100 million and 20 million.

In task terms, AutoRouter works out that at $0.184 per real bug fix, Pro’s $20 covers about 110 fixes at API prices, Max’s $100 about 540, and Ultra’s $200 about 1,090. These show what the same money buys on the API. They don’t show what a plan holds, because Anthropic doesn’t publish plan allowances. Third-party tests suggest a plan can hold many times its price in API-priced usage, but those are outside estimates.

Here is that arithmetic as monthly API cost at $0.184 a task, an illustration built from the figure above.

Tasks a monthAPI cost at $0.184 each
20About $3.68
110About $20.24
400About $73.60
1,000About $184

The rule of thumb for Claude Code pricing follows from that. If you run a handful of tasks a week, an API key is probably cheaper: at roughly $0.25 a task, 20 tasks cost about $5. If you work in Claude Code daily, the flat price is usually better, since Anthropic’s own budgeting guidance is about $13 per developer per active day, or $150 to $250 a month, across enterprise deployments at API prices. Earlier versions of those docs said $6 a day, so check which version you read.

Gotchas That Raise the Bill

Most surprise bills under Claude Code pricing come from a few specific settings.

  • A stray API key. If ANTHROPIC_API_KEY is set, Claude Code uses it instead of your subscription and bills API rates. Run /status to see which credential is active.
  • API credits that don’t cover Claude Code. Max and Ultra include $100 and $200 a month of Claude Platform API credits, but those are for building your own apps. They don’t apply to interactive Claude Code.
  • Agent teams. Anthropic’s docs say a team uses roughly 7 times the tokens of a standard session when teammates run in plan mode.
  • Cold caches. After a break longer than the cache lifetime, the next message reprocesses the whole context at uncached rates.
  • Long sessions. Each turn re-reads the conversation so far. Clear the session when you switch topics.

Claude Code vs. Cursor: Two Different Meters

Claude Code and Cursor charge for similar work in different ways. On a plan, Claude Code draws from limits Anthropic doesn’t size in tokens, and extra use bills at API rates. Cursor’s current plans are Free, Plus at $8 (Grok models only), Premium at $30, Super at $100 and Ultra at $200. From Premium up, Claude and GPT models are billed at API list price, and the allowance refreshes weekly.

Past each plan’s allowance, both bill Claude models per token at list price. What differs is what the plan fee buys: a pool of limits in Claude Code, a dollar allowance in Cursor. They’re also different tools. Claude Code is a terminal agent, and Cursor is an editor. Pages comparing them by seat price alone, including some that still list Cursor at $20, predate Cursor’s October change. Our guide to Cursor alternatives covers the wider field.

Claude Code Pricing for Teams

Team seats are for 2 to 150 people. They add central billing, SSO and admin controls. Finout’s example shows how mixed seats work. Ten Standard seats cost $200 a month on annual billing. Swap two or three of them for Premium seats at $100 each, and the bill rises to about $360 to $440. Ten Premium seats cost $1,000 a month, against $2,000 for ten individual $200 plans.

Admins can set spending caps at the organization and user level. Team plans also include pooled monthly API credits, capped at $500, for building on the Claude Platform. As with Max and Ultra, those credits don’t pay for interactive Claude Code. Enterprise costs $20 a seat on annual billing plus usage at API rates. It suits organizations that need SCIM, audit logs and custom retention.

Which Claude Code Plan Should You Pick?

Match the plan to how often you use it, since Claude Code pricing rewards steady daily use.

  • A few tasks a week: an API key with a spending limit, or Pro if you also use Claude chat.
  • Daily use, one developer: start on Pro, and move to Max only if you hit limits repeatedly.
  • All-day agent work: Ultra, which has the highest weekly usage and no five-hour limit once verified.
  • A team: Team seats, with Premium seats for the few people who hit limits. Enterprise adds SCIM and audit logs.
  • Automation or CI: an API key, with a workspace spending limit.

Whichever you choose, set a monthly limit on usage credits before you need one. Test anything an agent changes on a staging copy first.

Where Claude Code Pricing Is Heading

Plan names, credits and model prices all moved within weeks. Max 20x became Ultra in late October, Claude Fable 5.2 launched on November 9 at $8 and $40 per million tokens, and credits bought from December 7 will expire after 12 months. Expect more change. Re-read these numbers before you commit to an annual plan, and measure a week of your own usage first. If you’re comparing other metered builders, our guides to Replit pricing, the best vibe coding tools and the best AI app builders cover how each one meters usage.

Frequently Asked Questions

How much does Claude Code cost per month?

Claude Code pricing starts with a plan: Pro at $20 a month ($17 annually), Max at $100, or Ultra at $200. Team seats run $25 to $125 a month. With an API key you pay per token instead. There’s no standalone Claude Code price.

Is Claude Code free?

No. The free Claude plan doesn’t include Claude Code. You need a paid plan, a Team or Enterprise seat, or a Console account billed per token. Pro billed annually is the cheapest way in at $17 a month.

Does Claude Code cost extra on a Claude plan?

No, it’s included and shares the plan’s limits. You pay more only if you turn on usage credits, billed at API rates, or if an API key in your environment overrides your plan login.

What happens when I hit the Claude Code usage limit?

You can wait for the five-hour or weekly reset, upgrade, or turn on usage credits, which bill at standard API rates up to a monthly limit you set. Claude chat and Claude Code share the same limits.

Which model does Claude Code use by default?

Claude Opus 5.5, according to Claude Code’s model docs as summarized in October, at $4 and $20 per million input and output tokens. You can switch to Sonnet 5.5 or Haiku 5.5 for routine tasks to lower the cost.