Claude Code pricing in plain English. Pro $20, Max 5x $100, Max 20x $200, plus API. What you actually get on each tier and which one I run daily.
Claude Code is included in every paid Claude subscription. The CLI itself is free to install. The cost comes from the Claude plan that powers it: Pro at $20 a month, Max 5x at $100 a month, Max 20x at $200 a month, or pay per token through the Anthropic API. Team and Enterprise sit on top of that for org-wide rollouts. The plan you pick decides which model you run, how much usage you get per week, and whether you can crank effort up to xhigh on Opus 4.7.
I'm Tom. I run Claude Code on Max 5x most days and have stress-tested API billing for bursty agent work. I've also lived through the April 2026 pricing scare, when Anthropic briefly removed Claude Code from Pro for a slice of users before reversing the decision. This post is the version of pricing I wish someone had handed me before I made the wrong choice twice. Specific numbers, real receipts, and a calculator to help you pick a plan instead of guessing.
Claude Code costs $0 to install and from $20 a month to use seriously. The cheapest paid path is Claude Pro at $20 a month (or $17 a month billed annually). The most expensive consumer tier is Max 20x at $200 a month. Anywhere in between, you can switch to the Anthropic API and pay per token, which makes sense for spiky workloads. There is no separate Claude Code license. You pay for Claude, and the CLI is included.
Here is the at-a-glance table I keep in my notes.
Free: $0. Sonnet only, capped chat use, no production Claude Code work.
Pro: $20 a month, or $17 a month annual. Sonnet on Claude Code, modest weekly cap, default high effort.
Max 5x: about $100 a month. Sonnet plus Opus, roughly five times Pro's usage cap, xhigh effort unlocked on Opus 4.7.
Max 20x: about $200 a month. Same models, roughly twenty times Pro's usage, the highest priority queue, agent teams without flinching.
Team: $25 a user a month, or $20 a user a month annual. Pro-tier usage with shared billing, SSO, and admin controls.
Enterprise: call us. Custom SSO, audit logs, managed policies, and pooled usage.
API: pay per token. Sonnet runs $3 per million input tokens and $15 per million output. Opus 4.7 runs $15 per million input and $75 per million output. Prompt caching and batch billing knock those numbers down for repeat work.
Pro is the entry plan. Max is the same plan with much higher usage caps and Opus 4.7 access for Claude Code. Pro gives you Claude Code on Sonnet with a modest weekly cap. Max gives you Sonnet plus Opus, five or twenty times the cap, and the xhigh effort tier on Opus. Anthropic's pricing page lays out the headline numbers, but the day-to-day difference is which model you can actually run for hours, not which logo is on your invoice.
On Pro, every Claude Code session runs on Sonnet by default. You can flip to Opus inside a Pro Claude Code session for short bursts, but the usage cap on Pro hits fast and Opus eats it three to five times faster than Sonnet. On Max, you can sit on Opus all day. That alone is worth the upgrade if your work is the kind of multi-file refactor or ambiguous brief where Opus's reasoning earns its keep.
Claude Max comes in two flavours. Max 5x is about $100 a month and gives you roughly five times Pro's usage cap. Max 20x is about $200 a month and gives you roughly twenty times Pro's usage cap. Both unlock Opus 4.7, xhigh effort, and the priority lane that keeps you out of the queue when traffic spikes. The $100 jump from Pro to Max 5x is the single biggest quality-of-life upgrade in Claude Code. The $100 jump from Max 5x to Max 20x is for people running agent teams or multi-hour parallel sessions.
Use this decision matrix to pick a plan in under a minute. The honest tax bracket is built around the only two questions that matter: how often do you run Claude Code, and how much do you care about Opus.
Curious tinkerer, under five hours a week, mostly content and one-off scripts: Free or Pro. Pro is the right answer if you want to actually finish anything.
Solo builder, five to fifteen hours a week, building real projects: Pro if you can stay disciplined on Sonnet, Max 5x if you cannot. I recommend Max 5x to almost everyone in this bracket because hitting the Pro cap mid-build is more expensive in lost momentum than the $80 difference.
Daily operator, fifteen to thirty hours a week, agent teams and parallel sessions: Max 5x if you ration Opus, Max 20x if you do not. This is where I sit. Max 5x covers it for most weeks. The weeks I push harder, I bump effort down rather than upgrade.
Hardcore agent team, thirty plus hours a week, running Claude Code on the web and in the terminal in parallel, building products: Max 20x. Below that, you will hit the cap on Tuesday afternoon and spend the rest of the week regretting it.
Team of three to ten devs sharing one bill: Team plan at $25 a user, $20 if you pay annually. Pro-tier usage per seat with one invoice, SSO, and the ability to swap members in and out without billing chaos.
Bursty workloads with quiet weeks: API. If you run hot for three days then go quiet for ten, pay-as-you-go on the API often beats any subscription. Build a small budget alarm into your dashboard and you will not get caught out.
Yes for daily Claude Code users, no for casual ones. Max 5x at $100 a month is worth it the moment you start hitting Pro's weekly cap or wanting Opus on tap. Max 20x at $200 is worth it once you are running agent teams, parallel sessions, or building a product the agent ships work on. If you are still trying Claude Code out, Pro is the right starting point. The upgrade path is two clicks when you outgrow it.
Here is the test I use. Open /usage in Claude Code on a normal week. If you are above 60% of your weekly cap by Wednesday, Max 5x will pay for itself in saved time. If you are above 80% of Max 5x's cap by Wednesday, Max 20x will pay for itself in saved time. The math is rarely about the dollars. It is about the cost of stopping mid-build because the agent ran out of room.
Only if you are billing Claude Code through the API. By default, Claude Code uses your subscription quota (Pro, Max, Team), not your API credit balance. If you log in with claude login on a Pro or Max account, every token comes out of your weekly subscription cap. If you set ANTHROPIC_API_KEY in your shell or run claude login --api, every token comes out of your API credit balance instead.
This trips up almost everyone the first time. You buy $50 of API credit, never set the env var, and wonder why your balance never moves. The credit only burns when you tell Claude Code to use it. Run claude config get authMode to check which mode you are in. The two valid values are subscription and api.
Claude Code runs on a weekly usage cap that resets every seven days from your billing anchor. Anthropic does not publish the exact token figures and I will not pretend to either, because they have moved twice in the last six months. What stays constant is the multiplier between plans: Max 5x is roughly five times Pro, Max 20x is roughly twenty times Pro. Run /usage in Claude Code to see your current week's spend, your reset time, and which model you are burning through.
The official Claude Code documentation describes the rate limit as a rolling weekly window. In practice, that means a heavy weekend on Saturday and Sunday counts toward the same cap as your work on Monday and Tuesday. If you smash the cap, Claude Code degrades to a smaller model or asks you to wait until reset. It does not silently bill your card.
Effort does not cost extra dollars on a subscription. It costs tokens against your weekly cap. The /effort command lets you set the level for the current session: low, medium, high, xhigh, or max. Higher effort means more extended thinking on every turn, which means more tokens per response, which means you hit the cap faster. On API billing, higher effort directly increases your per-turn token spend.
In rough numbers, xhigh on Opus 4.7 burns about three to five times the tokens of high. Max effort can go higher again on hard reasoning tasks. The right move is to keep your default at high and reach for xhigh only when the work is genuinely hard. Plan mode plus xhigh is the combination I reserve for refactors that touch more than ten files.
Two steps. Buy a Pro or Max subscription on claude.com, then change your auth mode in the CLI. Open Claude Code, run /logout, then /login, and pick the subscription option when prompted. The CLI swaps from API to subscription billing immediately. Any unused API credits stay on your account and do not expire for twelve months.
If you want to make the switch programmatic, set the env var explicitly. CLAUDE_AUTH_MODE=subscription in your shell will pin Claude Code to your subscription regardless of the API key in your environment. Drop that into your .zshrc or .bashrc and you will never burn credits by accident again.
Pick Pro if Claude Code is a side tool you reach for a few hours a week. Pick Max 5x if Claude Code is part of your daily work. Pick Max 20x if Claude Code is the main way you ship work. The honest line between Pro and Max 5x is whether you want Opus 4.7 on tap. If yes, Max 5x. If no, Pro is fine for months.
Most people regret going to Max too late, not too early. The reason is that Pro's cap is generous on paper but feels small the moment you start running agents in the background or kicking off plan mode on a real refactor. The first week you hit the cap mid-build and wait two days for it to reset, you will know it is time to upgrade.
At the same tier, they are basically identical. Pro is $20 against Cursor Pro at $20. Max 20x is $200 against Cursor Ultra at $200. Cursor has a $60 Pro+ tier with no Claude Code equivalent, and Claude has Max 5x at $100 with no Cursor equivalent. The decision is not price. It is what each tool actually does, which I cover in the deep comparison below.
If you are deciding between the two, my full Claude Code vs Cursor breakdown walks through twelve real projects in both tools. The short version: Claude Code wins on agentic work, Cursor wins on inline autocomplete, and the smart move is to run them together for $40 a month total at the entry tier.
In April 2026, Anthropic ran a quiet test that removed Claude Code from the Pro tier for about 2% of users. The Register covered the change within hours of users noticing. The pushback was loud, and Anthropic reversed it inside a week. If you saw the news and assumed Pro no longer covers Claude Code, that was true for a sliver of users for a few days, and is no longer true for anyone.
The bigger lesson is that pricing here is still being calibrated. Anthropic ships Claude Code on subscription billing as a strategic bet, not a profit centre. They want serious builders inside the surface and will move tiers around to find the right line. Plan for one to two pricing changes a year and it will not catch you off guard.
Pro at $20 if you are starting out and want to test Claude Code on real work for a month before committing. Max 5x at $100 if Claude Code is in your daily stack. Max 20x at $200 if it is your primary surface for shipping software. Team if you are three or more devs sharing one bill. Enterprise if you have an IT team negotiating contracts. API if your usage is bursty enough that a flat monthly subscription does not fit.
My personal answer: Max 5x for almost everyone reading this. The tier is the best value in the lineup, the upgrade from Pro is the biggest quality-of-life jump, and the gap to Max 20x is mostly relevant once you are running parallel agent teams. Start there, watch /usage for a month, and adjust.
The Claude Code CLI is free to install. Real usage requires a Claude account, and the free tier on Claude is too capped to do meaningful work. Pro at $20 a month is the realistic floor. The full breakdown of what "free" actually gets you lives in my deep dive on whether Claude Code is free.
Yes. Pro at $20 a month includes Claude Code on Sonnet with a modest weekly usage cap. Anthropic briefly removed Claude Code from Pro in April 2026 for a small slice of users, then reversed the change. As of today, every Pro account gets Claude Code.
Pro lets you flip to Opus inside a session, but Pro's weekly cap was sized around Sonnet. Real Opus work eats it three to five times faster. If you want Opus on tap without watching the cap, Max 5x at $100 a month is the right tier.
Yes. Anthropic offers a student discount on Pro through .edu email verification. The exact percentage has shifted over the year, but expect roughly 50% off Pro for verified students. Max tiers are not currently part of the student program.
Pro is a flat monthly subscription with a weekly usage cap. The API is pay-per-token with no cap. Pro is cheaper for steady daily use. The API is cheaper for bursty workloads. Both run the same Claude models. The difference is billing, not capability.
Opus 4.7 is $15 per million input tokens and $75 per million output tokens at standard pricing. Prompt caching reduces input cost by up to 90% on cached reads. Batch billing through the Anthropic batches API runs roughly half price for non-time-sensitive jobs.
Cursor Ultra is $200 a month with roughly 20x Cursor Pro's usage. Claude Code Max 20x is $200 a month with roughly 20x Pro's usage. The headline price is identical. The difference is the tool around the model. Cursor Ultra is an AI-first IDE. Claude Code Max 20x is a multi-surface agent that runs in your terminal, IDE, desktop, browser, and phone.
For three to ten people, yes. Team at $25 a user (or $20 a user annual) gives you Pro-tier usage per seat with one bill, SSO, and admin controls. Below three people, individual Pro subscriptions are cheaper. Above ten, Enterprise starts to make sense for the audit logs and managed policies.
Almost certainly. Anthropic adjusted pricing twice in the last year, ran the Pro test in April, and is still calibrating where Claude Code sits in the lineup. Plan for one or two changes a year and check the pricing page when something feels off. The CLI itself does not change. The plan tiers around it do.
Pricing is the easy part. Picking a plan takes ten minutes. Actually using Claude Code well takes a weekend. The fastest way in is the free Claude Code Blueprint. Sixty minutes, no coding required, and you walk out with a working setup, your first build, and a clean answer to which tier fits your workflow. If you want a longer reference for the tool itself, the complete Claude Code guide covers every concept the pricing page assumes you already know.
If you want the version where you actually ship something in a month, the 30-Day Claude Code Challenge is the cohort. Four live calls, a real build per week, and a deployed agent on a real schedule by the end. By the time you are done, you will know your real Claude Code bill, your real plan, and your real workflow. Not the hypothetical version on a comparison post.
Five interactive lessons. Install Claude Code, build your first automation, and deploy it live on the internet — all in under an hour. Free, no coding required.
Grab the Blueprint →