Claude Code Pricing: Free or Paid? Plans, API and Real Cost

I use it every day in client work, on code as well as SEO scripts. The prices below were taken from Anthropic's official pages on October 7, 2026. They change often: check the date before you build a budget on them.

Claude Code pricing: the plan table in October 2026

There is no separate "Claude Code" subscription. You pay for access to Claude, and Claude Code comes with it, in the terminal, in VS Code, in JetBrains and in the desktop app. The only real decision is how much usage you need.

Public prices in US dollars excluding taxes, taken from claude.com; conversion into your currency only shows up at checkout. For the details of the plans on the chat side (projects, memory, models), see my article on Claude subscription pricing.

Pro: the entry ticket

Pro is the first tier that unlocks Claude Code. The annual commitment comes to $200 paid upfront, a $40 saving over twelve months. For a freelancer who codes a few hours a week, it's the sensible starting point. Note: the Pro subscription does not include API usage through the Console, which is paid separately.

Max: for daily use

Max comes in two tiers, $100 and $200 a month, for five and twenty times Pro's usage, with priority access to new features. There is no annual rate. It's the plan I recommend as soon as Claude Code becomes a daily work tool rather than an occasional extra.

Team and Enterprise: for a team

Team is counted in seats, from 2 to 150, and every seat gives access to Claude Code. Mixing the two seat types pays off: a few Premium seats for people who code all day, Standard seats for people who write or analyze. Enterprise works differently: a seat at $20 a month billed annually, plus actual consumption at API rates. An organization that budgets Enterprise as a fixed line item gets it wrong: consumption has to be modeled separately.

Is Claude Code free? What's possible and what isn't

Claude's free plan gives access to chat on the web, mobile and the desktop app, with limited volume. It does not give access to Claude Code: in the desktop app, clicking the Code tab prompts you to upgrade to a paid plan, and the official documentation states it in black and white. If what you mostly need is to chat with Claude, my article on Claude AI free details what the free plan allows.

Four doors remain open to use Claude Code without paying, each with its limits.

1. A 7-day invitation (guest pass)

Eligible Pro and Max subscribers have a limited number of invitations. The invited person gets seven free days of the Pro plan, Claude Code included. The condition: never having had a paid Claude subscription. On the subscriber's side, the /passes command in Claude Code shows the available invitations and the link to share. It's the simplest way to test the tool on a real project.

2. The API credit for new accounts

Anthropic says new API users receive "a small amount of free credits" for testing. The amount is not published. Once Claude Code is connected to that Console account, the credit is enough for a few trial sessions, not for real work.

3. The Claude for Open Source program

Anthropic offers six months of Max 20x to open source contributors who meet one of its criteria: maintaining a library that at least 500 repositories or 100 packages depend on, or that totals 200,000 downloads a month; being part of the core team of a recognized project (CPython, Node.js, Kubernetes, the Linux kernel…); having had at least 100 pull requests merged into third-party repositories within a year. At the end of the six months, the free subscription stops.

4. Claude Code with an open source model (Ollama)

This is the method that circulates under the headline "Claude Code free and unlimited". Ollama can run Claude Code with open source models installed on your machine: the ollama launch claude command handles it, and Ollama recommends a context window of 64,000 tokens or more. The cost is zero, but it's no longer Claude doing the work: you keep the Claude Code interface with another model, clearly less capable on long tasks, and you need a powerful machine. Useful for learning the tool, not for judging what it's worth.

What does Claude Code cost on the API?

With a Claude Console account, Claude Code is billed per token consumed, with no subscription and no minimum. Here are the rates of the current models, in dollars per million tokens, as of October 7, 2026:

Three mechanisms move the bill far more than the rate card itself.

Caching. Claude Code sends the whole conversation with every request, but it relies on the cache automatically: the part it has already seen is read back at one tenth of the input price (one twentieth on Opus 5.5). Writing to the cache costs 1.25 times the input price for five-minute retention, twice the input price for one hour. That's why a long break is expensive: the first request after the cache expires pays for the whole context again.

Model choice. Between Haiku and Opus 5.5, the gap is a factor of four. The Claude Code documentation is blunt about it: Sonnet is enough for most tasks, Opus should be kept for architecture decisions and multi-step reasoning. The /model command switches models mid-session.

Reasoning. Thinking tokens are billed as output tokens. On Opus 5.5, Sonnet 5.5 and Fable, thinking cannot be turned off; you can lower the effort level with /effort.

The tokenizer trap

Comparing two models on the price per million tokens is not enough. Since Claude 4.7, the models use a tokenizer that splits text more finely: for the same text, it produces about 30% more tokens, according to Anthropic's documentation. A recent model listed at the same rate as an older one therefore uses more units for the same work. The only serious comparison is one run on your own content.

What can be added on top

• Fast mode on Opus 5.5 doubles the rate: $8 input and $40 output per million tokens.
• US-only inference, for compliance reasons, applies a 1.1 multiplier to every line item.
• Web search through the API costs $10 per 1,000 searches, on top of the tokens from the pages it returns.

How much does Claude Code really cost per month?

The Claude Code documentation gives a rare benchmark, measured on enterprise deployments billed by usage: about $13 per developer per active day, and $150 to $250 per developer per month. For 90% of users, spending stays under $30 per active day.

Set against the subscriptions, the math speaks for itself. A developer who works with Claude Code every day uses roughly the price of a Max subscription on the API. The subscription then brings predictability, and covers Claude in the browser as a bonus. Conversely, a few hours of use a week often stays under Pro's $20.

To track your own usage, the /usage command shows, in a session, the estimated cost (on an API account) or your limit gauges (on a subscription), with the share used by skills, subagents and MCP servers.

Run the numbers for your case: enter your monthly usage, or your number of days of use, and the tool tells you which plan works out cheapest.

Subscription or API: how to choose

The tipping point is something you calculate, not something you guess. Here is the method I use.

1. Measure your real usage over a month: /usage in a session, or the Console's Usage page if you're on the API.
2. Convert it into API cost with the rate card above, cache included, or with the tool.
3. Compare it to the tiers of $20, $100 and $200. Below the tier that covers you, the subscription wins, and not only on price: it gives you a fixed amount and covers Claude in the browser.

Two subscription mechanics are worth knowing before you decide. First, the limits are not a monthly counter: they reset in five-hour windows, with a weekly cap on top. A heavy day can therefore stop you for a few hours even though your month has barely started. Second, Claude usage and Claude Code usage draw from the same counter: a long writing session in the browser reduces what you have left for coding. Once you hit the limit, you can wait for the reset, or turn on usage credits, billed at API rates.

Three signals point to the API, whatever the math says:

• Claude Code runs in scripts or a continuous integration pipeline, with nobody in front of the screen.
• You want a spending cap per workspace, which you can set in the Console.
• There are several of you and everyone's usage is very different, which is easier to model per token. Price out Team too before you multiply individual subscriptions.

How to cut your Claude Code bill

Cost follows context size: the more Claude Code rereads, the more you pay, in dollars on the API as in limits on a subscription.

• One session per task. /clear starts again from an empty context when you change topic; a session left open all day resends its whole history with every exchange.
• The right model. Sonnet by default, Opus when the task justifies it, Haiku for simple subagents.
• A short CLAUDE.md. The documentation recommends staying under 200 lines and moving specialized procedures into skills, loaded only when they're needed. The how-to is in Claude Skills.
• Fewer active MCP servers. Turn off the ones you don't use with /mcp; a command-line tool costs less context. See what an MCP server is.
• Precise requests. "Add validation to the login function in auth.ts" costs far less than "improve this project", which triggers an exploration of the whole repository.
• Plan mode before a big job: the agent proposes an approach before changing anything, so you don't pay for work that has to be redone.

What I pay myself

I use a subscription for day-to-day work, in the terminal as in the browser, and I open the API for everything automated: content pipelines, audit scripts, batch processing on thousands of URLs.

This split has a practical benefit. The subscription covers the unpredictable part, where I'm talking with the model and don't want to watch a meter. The API covers the industrial part, the one I can measure, cache and run as batches. The budget becomes readable, and above all, something I can optimize. This is the kind of pipeline I set up in an AI automation engagement.

One last piece of advice, whatever you choose: set a spending limit on your API account from day one. A poorly scoped agent stuck in a loop can burn through in one night what you had planned for the month.

Written by

Anthony Courtin

SEO, GEO & AI automation consultant

I help SMEs and independent professionals with their visibility (Google, ChatGPT, Perplexity) and with the real-world use of AI at work. I run on-site sessions to train teams and rethink their processes. Everything I write here, I use in client work.

Follow me on LinkedIn My YouTube channel (in French)