Claude Code Pricing (2026): Plans, API Costs and Limits

Claude Code itself is a free download, but every prompt runs on Claude models, and someone pays for those tokens. This guide explains the two ways to pay (a subscription or per-token billing), what each plan costs, how the usage limits work, and how to see and cut your spend.

All prices below were checked on October 24, 2026 on claude.com/pricing, the Claude Help Center and the Claude API pricing page. Prices are in USD, exclude tax and change often, so confirm on the official pages before you buy.

The two ways to pay for Claude Code

	Subscription	Per-token (API)
Accounts	Pro, Max, Team, Enterprise on claude.ai	Claude Console, or Amazon Bedrock, Google Cloud’s Agent Platform, Microsoft Foundry
What you pay	Fixed monthly fee (per seat for teams)	Every input and output token at the model’s list price
Limits	Five-hour session window plus a weekly limit	Your organisation’s rate limits and spend caps
Shared with chat	Yes, same allowance as Claude.ai	No, separate from any subscription
Best for	Individuals and teams who code with Claude daily	Automation, CI, spiky usage, or companies on a cloud contract

The free Claude.ai plan does not include Claude Code. The official setup docs state it plainly: Claude Code requires a Pro, Max, Team, Enterprise or Console account, or a third-party provider. For what each plan includes beyond coding, see our Claude pricing overview.

Subscription plans that include Claude Code

Plan	Price (as of Oct 24, 2026)	Usage	Claude Code
Free	$0	Basic	Not included
Pro	$20/month billed monthly, or $17/month billed annually ($200 up front)	Baseline	Included
Max 5x	$100/month	5x more usage per session than Pro	Included
Max 20x	$200/month	20x more usage per session than Pro	Included
Team, Standard seat	$25/seat/month monthly, $20 annually	More usage than Pro	Included
Team, Premium seat	$125/seat/month monthly, $100 annually	5x the Standard seat	Included
Enterprise	Listed as US$20/seat/month billed annually, 20-seat minimum; “usage cost scales with model and task”	Per contract	Included

Notes from the official pages:

• Max is monthly only. The Help Center says Max is “currently available as a monthly subscription only”, and upgrading between Max tiers is prorated.
• Team seats come in two tiers. Each member’s Claude Code usage draws from a per-seat allowance that resets on a rolling five-hour window and a weekly window, shared with Claude chat and Cowork.
• Enterprise pricing is seat plus usage. Talk to Anthropic sales for real numbers; the public listing is a starting point.
• Max’s monthly API credits don’t apply to Claude Code. The Help Center notes they “can’t be used in Claude Code or the Claude apps.”

How the usage limits work

On Pro and Max there are two limits:

1. A five-hour session limit, a rolling window that resets five hours after it starts.
2. A weekly limit on top of it. Max plans also have a separate weekly limit for Fable models.

Both are shared between Claude.ai and Claude Code. A long research chat in the browser eats into the same allowance as your coding session. Anthropic does not publish either limit in hours or tokens; Claude usage limits explained collects every documented rule per plan, and the Claude limits changelog tracks how they have changed.

When you hit a limit, Claude Code tells you which one (“You’ve hit your session limit” or “You’ve hit your weekly limit”) and when it resets. Switching models with /model won’t help for those, because they apply across all models. A model-specific message such as “You’ve hit your Opus limit” is different: switching to a model outside that family keeps you working.

Your options at that point:

• Wait for the reset. Recent versions can wait and continue the interrupted task automatically.
• Turn on usage credits and keep working at extra cost. Run /usage-credits after signing in with your subscription; on Pro and Max it opens Settings > Usage on claude.ai.
• Upgrade from Pro to Max, or Max 5x to 20x.
• Switch to a Console account and pay per token.

API (per-token) pricing

If you sign in with a Claude Console account, or run Claude Code on Bedrock, Google Cloud or Foundry, you pay per token. Anthropic’s list prices for the current models on the Claude API, as of October 24, 2026 (our Claude API pricing guide adds caching, batch discounts and worked examples):

Model	Input	Output	Cache hits
Claude Fable 5.1	$10 / MTok	$50 / MTok	$0.25 / MTok
Claude Opus 5.5	$4 / MTok	$20 / MTok	$0.20 / MTok
Claude Sonnet 5.5	$2 / MTok	$10 / MTok	$0.10 / MTok
Claude Haiku 5.5 (prompts up to 100k tokens)	$0.10 / MTok	$0.50 / MTok	$0.01 / MTok

MTok means one million tokens. Cloud providers publish their own prices for the same models.

Three things make Claude Code cheaper than the raw table suggests, and one makes it more expensive:

• Prompt caching is automatic. Claude Code re-sends your conversation on every request, but most of it is read from cache at a fraction of the input price.
• Model choice matters most. In Claude Code, opus currently resolves to Opus 5.5 and sonnet to Sonnet 5.5 on the Anthropic API. Anthropic’s own costs guide recommends Sonnet for most coding and Opus for complex architecture work.
• Thinking tokens are billed as output. Extended thinking is on by default and can add tens of thousands of tokens per request. Lower it with /effort on simple tasks.
• Agentic loops multiply requests. Every tool call sends another request carrying the conversation so far.

What does it cost in practice?

Anthropic’s costs documentation gives one real-world anchor: across enterprise deployments, the average is around $13 per developer per active day and $150–250 per developer per month, with costs below $30 per active day for 90% of users. Your numbers depend on model, codebase size, and habits like running several sessions in parallel, so run a small pilot before budgeting.

For teams on the API, the same page recommends per-user rate limits that shrink as headcount grows, for example 200k–300k tokens per minute per user for 1–5 users and 10k–15k for 500+ users.

Subscription or API: which should you choose?

Not sure which side of the line you are on? The plan picker prices your hours, model mix and working style at API rates and compares the total with Pro, Max and Team.

Your situation	Better fit
You code with Claude most days	Pro, then Max when you keep hitting limits
You hit Pro limits several times a week	Max 5x
You run long agent sessions or parallel sessions all day	Max 20x
A team of 2+ wanting shared billing and admin controls	Team (mix Standard and Premium seats)
CI jobs, headless scripts, GitHub Actions	API key from the Console, with a workspace spend limit
Your company already buys AWS, Google Cloud or Azure	Bedrock, Google Cloud’s Agent Platform or Foundry

A common setup is a subscription for interactive work plus a separate Console key for automation, so CI never drains your personal allowance. Torn between the two subscription tiers? Our Claude Pro vs Max comparison works through it.

Avoid the surprise API bill

This catches people every week. If the ANTHROPIC_API_KEY environment variable is set and you approve it when Claude Code asks, Claude Code uses the key instead of your subscription, and you pay API rates. The Help Center flags this explicitly.

# Check whether a key is set in your shell
echo $ANTHROPIC_API_KEY

# Remove it for this shell, then sign in with your subscription
unset ANTHROPIC_API_KEY
claude
# inside the session:
/login

Also look for the variable in your shell profile, .env files loaded by direnv, and the env block of ~/.claude/settings.json.

Track your spend

/usage

Run /usage inside Claude Code. On a Pro, Max, Team or Enterprise plan it shows your plan usage bars plus a breakdown of what consumed them: skills, subagents, plugins and individual MCP servers as a share of the total. It also lists recommendations such as trimming an oversized CLAUDE.md, and in recent versions you can press f to have Claude Code apply the ones it can.

On API billing, the Session block shows an estimated dollar cost for the current session, computed locally at list price. For authoritative numbers, use the Usage page in the Claude Console.

Other tools

• Status line: show cost, context and your five-hour and weekly limit usage permanently; our statusline recipe has a tested script.
• /insights: an HTML report on how you work, written to ~/.claude/usage-data/report.html. It costs tokens to generate.
• ccusage: a community CLI that reads Claude Code’s local logs and reports token usage by day or session.
• --max-budget-usd: in print mode (claude -p), stops a run once estimated spend reaches a cap. Useful in CI.

Seven ways to spend less

1. /clear between unrelated tasks. Stale context is re-sent on every request. Use /rename first so you can /resume later.
2. Match the model to the job. Sonnet for routine work, Opus for hard design problems, model: haiku for simple subagents.
3. Keep CLAUDE.md under about 200 lines. It loads into every request; move specialised procedures into on-demand skills. Our CLAUDE.md guide shows how.
4. Lower effort for simple tasks with /effort.
5. Send noisy work to subagents. Test runs and log digging stay in the subagent’s context and only a summary returns. See the subagents guide.
6. Disable MCP servers you aren’t using via /mcp, and prefer CLIs like gh where they exist. The MCP servers guide covers setup.
7. Use plan mode for big changes (Shift+Tab), so you approve the approach before Claude spends tokens on the wrong one.

The context management guide goes deeper on /context, /compact and /clear, and how to reduce Claude token usage covers chat and API habits too.

Where pricing fits in your Claude kit

Budget is part of choosing your setup. Skills, hooks and subagents are free features; what costs money is the tokens they consume. If you are weighing a paid kit against building your own, our Agentkit.best vs DIY comparison and the Claude kit hub cover the trade-offs.