Claude Code pricing looks simple until you try to pick a plan: Pro, two Max tiers, Team seats, Enterprise, and the option to skip subscriptions and pay per token. This post lays out what each option costs as of October 2026, how the usage limits behind them work, which one fits which kind of developer and how to make whatever you pay go further.
Prices change. Everything below was checked against Anthropic’s official pricing page and the Claude Code costs docs in October 2026. Check those pages before you buy; prices shown there don’t include tax and may differ in the mobile apps.
Claude Code pricing at a glance
Claude Code is not sold on its own. You get it with a Claude subscription, or you pay for the tokens it uses through the API.
| Plan | Price (as of October 2026) | Claude Code | How usage is limited |
|---|---|---|---|
| Free | $0 | Not included | - |
| Pro | $20/month, or $17/month billed annually ($200 up front) | Included | 5-hour session window + weekly limit |
| Max 5x | $100/month | Included | 5× Pro usage per session + weekly limit |
| Max 20x | $200/month | Included | 20× Pro usage per session + weekly limit |
| Team, Standard seat | $25/seat monthly, $20/seat billed annually | Included | More than Pro, per seat |
| Team, Premium seat | $125/seat monthly, $100/seat billed annually | Included | 5× a Standard seat |
| Enterprise | $20/seat/month billed annually, 20-seat minimum, plus usage at API rates | Included | Usage billed per token |
| API (pay-as-you-go) | Per million tokens, by model | Works with an API key | Your spend limits |
Two things are easy to miss. First, the free Claude plan does not include Claude Code; installing the CLI costs nothing, but you can’t log in to it with a Free account. Second, Max is billed monthly only, and moving from Pro to Max is prorated.
Pro: the entry point
Pro is where most people start. As of October 2026 it costs $20 a month, or $17 a month if you pay $200 for the year up front. It includes Claude Code, the Claude apps and the other Pro features.
Pro gives you at least five times the Free plan’s usage per five-hour session, plus a weekly limit. In practice that is enough for a few focused sessions a day with one agent. It runs out quickly if you keep long conversations open, work in a large monorepo or run two agents at once.
Pro fits you if: you use Claude Code for part of your day, mostly one agent at a time, and you are fine waiting for the window to reset now and then.
Max 5x and Max 20x: for all-day use
Max comes in two tiers:
- Max 5x ($100/month): five times Pro’s usage per session.
- Max 20x ($200/month): twenty times Pro’s usage per session.
Both include higher output limits, early access to new features and priority access at busy times. Max is billed monthly only.
The jump from Pro to Max 5x is the one most developers think about. It makes sense when the limit is the thing slowing you down, not occasionally but most days. Max 20x is for people who run Claude Code all day, often with several agents in parallel.
As of October 2026 Max plans also come with monthly API credits for building on the Claude Platform ($100 on Max 5x, $200 on Max 20x). Claude Code itself draws on your plan’s usage limits, so check the pricing page for how the credits work before you count them in.
Max fits you if: you hit the Pro limit on most working days, run more than one agent at a time, or lose real time waiting for resets.
Team and Enterprise
Team
Team is for groups of 2 to 150 people and includes Claude Code on every seat. As of October 2026 there are two seat types:
- Standard seat: $20 per seat per month billed annually, $25 monthly. More usage than Pro.
- Premium seat: $100 per seat per month billed annually, $125 monthly. Five times the usage of a Standard seat.
Each seat has its own allowance with a rolling 5-hour window and a weekly window. That allowance is shared between Claude Code and the person’s use of Claude chat, so a heavy chat day leaves less for coding. A sensible setup is Standard seats for most people and Premium seats for the few who run agents all day.
Enterprise
Enterprise works differently. The seat costs $20 per month billed annually with a 20-seat minimum, and usage is billed on top at API rates. Cost therefore scales with the model and the work, not with a fixed tier. It suits larger organisations that want central controls and predictable governance more than a fixed monthly number per person.
Paying per token with the API
You don’t need a subscription at all. Claude Code can run with an API key from the Claude Console, and then it charges by token consumption. When you first authenticate, the Console creates a “Claude Code” workspace, and you can set spend limits on it.
API prices per million tokens as of October 2026:
| Model | Input | Output | Cache write (5 min) | Cache read |
|---|---|---|---|---|
| Opus 5.5 | $4 | $20 | $5 | $0.20 |
| Sonnet 5.5 | $2 | $10 | $2.50 | $0.10 |
| Haiku 5.5 (prompts up to 100K) | $0.10 | $0.50 | $0.125 | $0.01 |
Notice how cheap cache reads are compared with fresh input. Claude Code leans on prompt caching heavily, which is why long sessions cost far less than a naive token count suggests. A long break lets the cache expire, though, and the next turn pays full input price for the whole context again.
What does the API cost in practice?
Anthropic’s costs docs give a reference point: across enterprise deployments, the average is around $13 per developer per active day and $150 to $250 per developer per month, with costs below $30 per active day for 90% of users. Your own number depends on codebase size, model choice and how many agents you run.
That gives a rough rule. If your API spend would land well under $20 a month, the API is cheaper than Pro. If you work with Claude Code most days, a subscription is usually cheaper and certainly more predictable.
Claude Code also runs on Amazon Bedrock, Google Cloud and Microsoft Foundry, billed per token through those providers. That route mostly matters to companies that already buy cloud capacity there.
How Claude Code usage limits work
On a subscription you don’t pay per token, but your usage limits work much the same way: every token Claude Code reads and writes counts against them.
- Session window: usage is counted in a 5-hour window. When you hit the cap, you wait for the window to reset.
- Weekly limit: a second cap across all models, reset at a fixed time for your account.
- Model-specific limits: some plans have separate caps for particular models. Hitting one of those is the only case where switching models helps.
- Shared pool: every Claude Code session on the same account draws from the same allowance. On Team and Enterprise, so does Claude chat.
Run /usage inside Claude Code to see your plan’s usage bars, where the usage went and what to change. If you want to keep working past a limit, /usage-credits lets you set up extra usage with a monthly spend cap.
For the full picture, including what to do mid-task when you run out, see Claude Code usage limits.
Why usage climbs faster than you expect
The costs docs list a few causes that catch people out:
- The baseline context: the system prompt, CLAUDE.md, tool definitions and MCP servers are sent on every turn.
- Long conversations: every turn carries the history so far until you compact or clear.
- Cache misses after breaks: come back after lunch and the first turn re-reads everything at full price.
- Subagents and parallel agents: each one has its own context and all of them draw from the same pool.
- Agent teams: Anthropic notes that agent teams use roughly seven times more tokens than a standard session when teammates run in plan mode.
Which plan should you pick?
| You are… | Start with | Move up when |
|---|---|---|
| Trying Claude Code a few times a week | API with a spend limit, or Pro | You use it most days |
| Using one agent for part of your workday | Pro | You hit the session limit most days |
| Coding with Claude all day, often two agents | Max 5x | You still wait for resets every week |
| Running several agents in parallel all day | Max 20x | - |
| A small team | Team, Standard seats, a few Premium | Specific people keep hitting the cap |
| A large company with governance needs | Enterprise | - |
| Building your own tools and automations | API | - |
One test before upgrading: look at /usage for a week. If the limit is hit because of bloated context or the wrong model, an upgrade just lets the waste run longer. Fix that first, then decide.
How to keep Claude Code costs down
Most savings come from habits, not from the plan you’re on. These are the points the official costs docs emphasise, plus a few that follow from how billing works.
Pick the model for the job
The default model on Pro, Max, Team, Enterprise and the API is Opus 5.5. It is excellent for hard reasoning, but most everyday coding doesn’t need it. Use /model to switch to Sonnet for routine work and save Opus for architecture, tricky bugs and big refactors. For simple subagents, set model: haiku in their definition.
/effort matters too: lower effort means fewer thinking tokens on easy tasks.
Keep the context lean
- Run
/contextto see what’s taking space. - Keep CLAUDE.md under about 200 lines and move longer workflows into skills, which only load when needed. More in how to write CLAUDE.md.
- Disable MCP servers you aren’t using; where a CLI tool does the same job, prefer it.
- Use
/clearbetween unrelated tasks and/compactin long sessions.
The context window guide goes deeper on this.
Plan before you build
A wrong turn halfway through a big change costs tokens twice: once to build it, once to undo it. The costs docs recommend plan mode for complex tasks for exactly this reason. Claude explores and proposes a plan, you correct it, and only then does it write code.
Write specific prompts
“Improve this codebase” makes Claude read half the repository. “Add input validation to the login handler in auth.ts” lets it go straight to the right file. Name files, functions and the expected result.
Filter noisy output
Long test logs and build output fill the context fast. A hook that trims output before Claude sees it, for example showing only failing tests, saves tokens on every run. See the hooks guide.
Cap your spend
On the API, set a workspace spend limit in the Console. For scripted runs, --max-budget-usd stops a session at a fixed dollar amount. On a subscription, keep extra usage capped with a monthly limit if you enable it.
More detail in how to cut Claude Code token usage.
Keeping track of Claude Code costs in AgentVera
AgentVera is a desktop app for macOS, Windows and Linux that runs Claude Code and other CLI agents side by side, per project. It uses the Claude Code login you already have, so it works with whatever plan you pay for and doesn’t add a token bill of its own. A few features help directly with cost:
- Token budgets: set a cap per agent, per turn or per day under Settings → Orchestration → Token budget. AgentVera warns at 80% and stops the agent at the cap without losing its work. A chip in the header shows what you’ve spent so far.
- Automatic model: sends simple tasks to Haiku, normal ones to Sonnet and hard ones to Opus, so you don’t pay Opus prices for a rename.
- Token savings: lighter context profiles, auto-compact before the window fills, and compressed command output so test logs don’t eat your limit.
- Account switching: the status bar shows each account’s 5-hour and weekly usage. Where your provider’s terms allow more than one account, a chain moves agents to the next account when a limit runs low.
Token budgets, the automatic model and token savings are on the free plan. Account switching is on Pro and Team. See AgentVera pricing or download AgentVera.
Wrapping up
As of October 2026, Claude Code starts at $20 a month with Pro, steps up to $100 and $200 with Max, and comes as Standard or Premium seats on Team. Enterprise and the API bill per token instead. Start on Pro or a capped API key, watch /usage for a week, and fix context and model choices before you pay for a bigger plan. When you do upgrade, it should be because the work needs it, not because the waste does.