Notes7 min read

What Claude Code really costs: plans, limits and habits

You need a paid Claude plan, or an API key that bills per use. On a plan, usage limits decide how much you can do before you wait or top up, and choosing the right model and keeping your context lean make the same plan go much further.

From the course · lesson 0.12

How much does Claude Code cost? This note won't quote prices. They change, and an old number in a blog post is worse than no number. Instead, it explains how the pricing works, so you can read Anthropic's own pages with confidence and make your plan last.

How much Claude Code costs: two ways to pay

There are two ways to pay for Claude Code.

A Claude plan. You pay a monthly subscription and get a set amount of use. Claude Code is included in the paid plans: Pro, Max, Team and Enterprise. It isn't part of the Free plan. Your plan covers both the Claude app and Claude Code.

An API key. You pay for exactly what you use, billed per token through the Claude Console, Anthropic's developer site. This suits teams and automated jobs more than someone learning.

For most people starting out, a plan is simpler. You know the monthly cost up front, and the worst case is that you wait for your limit to reset.

How to read the pricing page

Open claude.com/pricing and look for three things:

  1. Which plans include Claude Code. It's listed in each plan's features.
  2. How much use each plan gives. The page describes this in relative terms. Max, for example, is described as a multiple of Pro's usage.
  3. The usage limits link. It explains how and when your use resets.

Limits in plain words

Anthropic's help center separates two kinds of limit, and the difference matters.

Usage limits are your budget over time. They decide how much you can do across all your conversations before you need to wait. On paid plans, use resets on a rolling session window, and there are weekly limits on top. Your budget is shared between the Claude app and Claude Code, so a long chat in the app counts too.

Length limits are about a single conversation. Each one has a context window, which is how much Claude can hold in mind at once. When it fills up, Claude Code summarizes the older parts so you can keep going.

Both come down to tokens. A token is a small piece of text, roughly part of a word. Here's the key fact: every time you send a message, Claude Code sends the whole conversation again, not just your new line. A long session costs more per message than a fresh one, even if you type the same thing.

When you get close to a limit, Claude Code warns you. When you hit it, you can wait for the reset, upgrade, or turn on usage credits. Usage credits let you keep working past your limit, and they're billed separately from your plan. You can set a monthly spend cap for them.

To check where you stand, type /usage in Claude Code. On a paid plan it shows your usage bars and a breakdown of what used them, such as skills, subagents and connectors. On claude.ai, the same picture lives under Settings > Usage.

Which model for which job

Claude comes in different sizes, called models. Bigger models reason more deeply, and each request uses more of your budget. Smaller ones are faster and cheaper. Think of it like picking a pen. You wouldn't sketch a wireframe with a technical pen, or sign a contract with a marker.

ModelGood forExample in Kudos
HaikuQuick, mechanical jobsRename a button label across the app
SonnetMost everyday buildingAdd a "copy embed code" button to the testimonial wall
OpusPlanning, tricky bugs, unfamiliar groundDecide how customer approvals should work before anything is built
FableThe hardest, longest multi-step workWork through a large change that smaller models struggle with

Type /model to see which models your plan offers and switch between them. Your plan comes with a default, and /model shows it. On some plans, the largest model draws on usage credits instead of your included limits, and the model picker labels this. In a normal session, Claude Code asks for your OK before it bills credits.

There's also a mixed option called opusplan. It uses Opus while you plan and switches to Sonnet when Claude starts building, so you pay for deep thinking only where it counts.

Effort, the second dial

Most models also have an effort level, which you set with /effort. Lower effort is faster and uses less. Higher effort makes Claude check more of its own work. The default suits most tasks.

Anthropic's own guidance offers a simple test when Claude gets something wrong. Did it lack knowledge? Try a bigger model. Did it not try hard enough, like skipping files or not running tests? Raise the effort.

Habits that keep sessions cheap

Think of context as a glass of water. Every message, file and tool result pours in more. Every time you send a message, the whole glass goes along with it. Most cost-saving habits are about keeping the glass from overflowing with things you no longer need.

Usage todayStandby
Many streams of work condensed into one line, so you can see what your plan is actually spent on.

Clear between tasks

When you switch to unrelated work, type /clear. It starts a fresh conversation with an empty glass. Old context from a finished task just adds cost to every new message. Anthropic's docs note that /clear itself costs nothing. Use /rename first if you want to find the old conversation later with /resume.

Compact when you want to keep going

If you're mid-task and the glass is getting full, type /compact. Claude summarizes the conversation so far and carries on from the summary. You can tell it what to keep: /compact focus on the approval flow decisions. Compacting reads the whole conversation to summarize it, so it's a big request in itself. Don't use it as a reflex.

Plan before you build

Press Shift+Tab to switch to plan mode. Claude explores and proposes an approach, and nothing gets built until you approve it. Anthropic's support article puts the math simply: "A plan costs a few hundred tokens." A wrong build you throw away costs far more.

Be specific

A vague request like "improve the app" makes Claude read everything. "Add a character limit to the testimonial form in the submit page" lets it go straight to the right file.

Stop early

If Claude heads the wrong way, press Esc. Then use /rewind to go back to an earlier point. Waiting for a wrong answer to finish only spends more.

Trim what's always loaded

  • Connectors. Each connected server adds a little to every session. Turn off ones you aren't using in /mcp.
  • CLAUDE.md. This file loads every session, so keep it to essentials. Move long, specialist instructions into skills, which load only when they're needed.
  • Subagents. These are helpers with their own context window. They keep noisy work, like reading logs, out of your main conversation. They still use your plan, so don't spin them up for small jobs.

Type /context to see what's taking up space right now.

Try it

Do this on your next session, in any project you have open.

  1. Type /usage and note where your bars are.
  2. Type /context to see what's already in the glass.
  3. Type /model and check your default.
  4. Switch to plan mode with Shift+Tab and ask for a plan for one small feature. Approve it, then let Claude build.
  5. When the feature works, type /clear before you start the next thing.
  6. Check /usage again tomorrow. The breakdown will show you where your plan really goes.

Lesson 0.12 in Chief of Agents covers models, limits and costs in your own project, with a plans and limits cheat sheet.

Sources

Learn this inside Claude Code.

Join the waitlist for Chief of Agents. You'll get the free beginner's handbook and skills right away, and hear first when the course opens.

One email to confirm, then the handbook and skills. Unsubscribe anytime.

Keep reading

  1. 8 min read

    Claude Code Projects: what they are and how to use them

    Claude Code Projects explained: one coordinator chat starts a thread per task, remembers your decisions, and works while you're away. How to start one.

  2. 6 min read

    Claude Code for designers: what it is, and why it isn't just for engineers

    What a coding agent does, why directing one is a design skill, and a safe first hour for designers new to Claude Code.

  3. 7 min read

    Agents, skills, MCPs and hooks, explained with Figma

    Agent vs skill vs MCP, plus subagents, hooks and plugins: seven Claude Code terms, each matched to something you know in Figma.

All notes →

© 2026 Chief of Agents · Notes · PrivacyIdeate. Build. Sell. You lead, the agents work.