Claude Code Pricing Guide 2026: Pro, Max, Team, API, and Break-Even Cost

“I already pay for Pro, so why did an API charge appear?” Claude Code pricing becomes confusing when subscription usage, extra usage, and API automation are treated as one wallet. They are separate billing paths, even when the same command-line tool is involved.

ClaudeCodeLab runs scheduled content operations as well as interactive repository work. Mixing those modes made it difficult to tell which allowance a run consumed. A pricing table alone did not solve that operational problem. The useful record is the authentication method, execution mode, model, limit result, and next billing action for each job.

This guide was verified on July 22, 2026 against the official Claude plans and pricing, Claude API pricing, Claude Code cost management, Pro and Max Claude Code billing guide, and current Agent SDK billing notice. Prices can change. Recheck the official pages immediately before purchasing.

Key Takeaways

• Interactive Claude Code in a terminal or IDE uses the allowance included with Pro, Max, Team, or Enterprise authentication.
• Extra usage enabled after a plan limit is pay-as-you-go and separate from the subscription price.
• The proposed monthly credit for claude -p and the Claude Agent SDK is paused and unavailable; subscription-authenticated use still draws from plan limits.
• Automation authenticated with ANTHROPIC_API_KEY should be budgeted as metered API usage.
• Change plans only after measuring seven days of limit interruptions, time saved, and review rework.

Separate the Three Billing Wallets First

Execution and authentication decide which wallet pays.

flowchart TD
  A[Use Claude Code] --> B{Authentication}
  B -->|Subscription OAuth| C[Pro / Max / Team allowance]
  B -->|API key| E[Metered Claude API billing]
  C --> D{Execution mode}
  D -->|Interactive terminal or IDE| I[Plan allowance]
  D -->|claude -p or Agent SDK| I
  C --> F{Plan limit reached}
  F -->|Wait| G[Allowance reset]
  F -->|Approve extra usage| H[Extra usage at standard API rates]

Interactive sessions authenticated with Pro or Max share usage with Claude on web, desktop, and mobile. There is a rolling five-hour allowance plus weekly limits. There is no reliable fixed message count because model choice, conversation length, files, and tools change consumption.

Anthropic announced a separate monthly credit for claude -p and Agent SDK use that was expected to start on June 15, 2026. A June 16 update paused that change. The credit is not currently available. With subscription OAuth, claude -p and Agent SDK use still draw from subscription limits; API-key authentication remains metered API billing. Do not put the historical $20, $100, or $200 proposal preserved lower on the Help Center page into a current budget.

Official Plan Baseline on July 22, 2026

Option	Published Price	Best Fit	Check Before Buying
Pro	$17/month equivalent billed $200 annually, or $20 monthly	Learning, light fixes, occasional reviews	How often real work stops in seven days
Max 5x	$100/month	Daily solo development and longer sessions	Whether interruption costs exceed $100/month
Max 20x	$200/month	Heavy multi-repository work	Whether Max 5x weekly limits threaten delivery
Team Standard	$20/seat/month annual, $25 monthly	2–150 people needing SSO and central billing	Whether admin controls justify a team plan
Team Premium	$100/seat/month annual, $125 monthly	Selected heavy coding users	Which users actually need five times Standard usage
Enterprise	$20/seat/month plus usage at API rates	SCIM, audit logs, retention, granular controls	Security requirements and metered budget
Claude API	Per-model token billing	CI, nightly jobs, measurable automation	Caps, alerts, retries, and stop conditions

Taxes, regions, and mobile app-store pricing may differ. Max is monthly only. Team organizations can mix seat types, so a measured pilot is cheaper than assigning Premium to everyone.

What Claude Code Can Decide and What a Human Must Decide

Work	Delegate to Claude Code	Human Decision
Usage log cleanup	Normalize task, duration, limit hit, and model into CSV	Monetary value of interrupted time
API estimate	Calculate monthly cost from input, output, and cache tokens	Required quality and acceptable model
Plan comparison	Normalize public prices into one table	Tax, contract, cancellation, and security terms
Cost reduction	Find oversized logs, stale context, and expensive model use	Which steps may use a lower-quality model
Team rollout	Rank candidate seat types from measured use	Who receives Premium and spending authority

“Choose the cheapest plan” is too broad. It omits delivery risk and review cost. Delegate calculation and candidate generation, but retain human ownership of business impact and contracts.

Use Case 1: A Solo Developer Measures Pro for Seven Days

Input: Task, start and end time, limit interruptions, and rework time.

Output: Weekly time saved, interruption cost, and the break-even point for Max.

Human check: Did the limit affect a deadline, or would smaller prompts and cleaner context solve it?

Start with Pro for bug fixes, small features, tests, and documentation. Consider Max 5x when useful work is stopped repeatedly and the value of waiting exceeds the $80 monthly difference. If the cause is an oversized log or stale conversation, fix Claude Code context and token use first.

Use Case 2: A Small Team Mixes Seat Types

Input: Active days, work type, limit hits, review rework, and required admin controls per member.

Output: Standard and Premium candidates plus a 90-day pilot budget.

Human check: SSO, offboarding, audit, data retention, and procurement requirements.

Team supports 2–150 people. Assign Premium to people doing migrations, long reviews, and large refactors; start everyone else on Standard. Combine the budget decision with the Claude Code permissions guide so higher usage does not mean broader uncontrolled access.

Use Case 3: Scheduled Jobs Use Agent SDK or API Budgets

Input: Runs per day, input and output tokens, cache usage, and retry count.

Output: Monthly estimate, cost per successful run, and a stop condition.

Human check: Should non-interactive work use subscription allowance or explicit API-key billing?

CI log summaries, dependency PR drafts, and ticket classification are measurable. Open-ended implementation is not. For scheduled operations, record start, finish, execution mode, model, and failure reason for every job. A retry loop without a budget cap can cost more than the successful work.

Runnable Claude API Cost Estimator

On July 22, 2026, Sonnet 5 has introductory pricing through August 31: $2/MTok input, $10/MTok output, $2.50/MTok five-minute cache writes, and $0.20/MTok cache reads. Standard rates from September 1 are scheduled to be $3/$15. Opus 4.8 is $5/$25 and Haiku 4.5 is $1/$5 for input/output.

// estimate-claude-api-cost.mjs
const rates = {
  sonnet5Intro: { input: 2, output: 10, cacheWrite: 2.5, cacheRead: 0.2 },
  sonnet5Standard: { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.3 },
  opus48: { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.5 },
  haiku45: { input: 1, output: 5, cacheWrite: 1.25, cacheRead: 0.1 },
};

const [model = "sonnet5Intro", days = "20", input = "0.8", output = "0.15",
  cacheWrite = "0.05", cacheRead = "0.4"] = process.argv.slice(2);
const rate = rates[model];
if (!rate) throw new Error(`Unknown model: ${model}`);

const usage = [days, input, output, cacheWrite, cacheRead].map(Number);
if (usage.some((value) => !Number.isFinite(value) || value < 0)) {
  throw new Error("Usage values must be non-negative numbers");
}

const [activeDays, inputMTok, outputMTok, writeMTok, readMTok] = usage;
const dailyUSD = inputMTok * rate.input + outputMTok * rate.output
  + writeMTok * rate.cacheWrite + readMTok * rate.cacheRead;

console.log(JSON.stringify({
  model,
  activeDays,
  dailyUSD: Number(dailyUSD.toFixed(2)),
  monthlyUSD: Number((dailyUSD * activeDays).toFixed(2)),
}, null, 2));

Run it:

node estimate-claude-api-cost.mjs sonnet5Intro 20 0.8 0.15 0.05 0.4

The sample returns $3.31 per active day and $66.10 for 20 days. Replace the sample token quantities with API response or billing data.

Calculate the Max Break-Even Point

Compare a fixed plan with the value of time it actually saves.

// plan-break-even.mjs
const [plan = "Max 5x", price = "100", hourlyValue = "50", savedHours = "3"] =
  process.argv.slice(2);
const values = [price, hourlyValue, savedHours].map(Number);
if (values.some((value) => !Number.isFinite(value) || value < 0)) {
  throw new Error("Price, hourly value, and saved hours must be non-negative numbers");
}

const [monthlyPriceUSD, hourUSD, hours] = values;
const breakEvenHours = hourUSD === 0 ? null : monthlyPriceUSD / hourUSD;
const savedValueUSD = hourUSD * hours;

console.log(JSON.stringify({
  plan,
  monthlyPriceUSD,
  savedValueUSD,
  netValueUSD: savedValueUSD - monthlyPriceUSD,
  breakEvenHours,
}, null, 2));

node plan-break-even.mjs "Max 5x" 100 50 3

At $50 per hour, Max 5x breaks even after two saved hours per month. Subtract time spent reviewing output, fixing wrong changes, and recovering from failures.

Pitfalls: Six Ways the Bill Stops Matching Expectations

1. ANTHROPIC_API_KEY Is Still Present

Cause: An approved API key can take precedence when you intended to use subscription authentication.

Fix: Check /status before long work. To return to subscription login, remove the variable (Remove-Item Env:ANTHROPIC_API_KEY in PowerShell or unset ANTHROPIC_API_KEY in bash) and use /login again.

2. /usage Dollars Are Treated as the Invoice

Cause: The session dollar amount is a local estimate for API users. It is not a Pro or Max subscription invoice.

Fix: Use plan bars for subscription allowance and Claude Console as the authoritative API bill.

3. The Paused Agent SDK Credit Is Put Into the Budget

Cause: The lower part of the official page preserves the pre-launch plan, while the current banner says the change is paused and the credit is unavailable.

Fix: Do not treat the historical amounts as a balance. Label every job interactive, agent-sdk-subscription, or api-key.

4. Annual Equivalent Is Treated as Monthly Cash Flow

Cause: Pro’s $17 is the monthly equivalent of $200 billed up front, not a $17 monthly charge.

Fix: Keep upfront cash and monthly equivalent in separate budget columns.

5. Token Counts Are Assumed Equal Across Models

Cause: Anthropic documents that the newer tokenizer used by Sonnet 5 and Opus 4.7+ produces about 30% more tokens for the same fixed text than the previous tokenizer, with workload variation.

Fix: Benchmark each candidate model with the same representative input instead of comparing list prices alone.

6. Every Limit Hit Triggers an Upgrade

Cause: Huge logs, unrelated files, stale conversations, and broad tasks can create avoidable usage.

Fix: Filter input and split tasks first. Upgrade when seven days of measured work still shows delivery interruption.

The 15-Minute Weekly Decision Loop

1. Use /status for authentication and /usage for allowance.
2. Record task, model, limit hit, time saved, and rework in one row.
3. For automation, aggregate runs and input, output, cache-write, and cache-read tokens.
4. Run both calculators to update API cost and plan break-even.
5. Separate workflow waste from genuine allowance shortage.
6. Recheck official prices immediately before changing a contract.

New users should start with the Claude Code getting started guide. For deeper metered-cost mechanics, read the LLM API cost guide.

Put Cost Controls Into Your Repository

The Claude Code Setup Guide is the primary next step when you need authentication, permissions, CLAUDE.md, hooks, and CI controls in one implementation path. It is an English-language paid guide and helps prevent API-key mixing and missing verification rather than stopping at a pricing comparison.

What We Actually Tested

We ran the API estimator with Sonnet 5 introductory rates, 20 active days, 0.8 MTok input, 0.15 MTok output, 0.05 MTok cache writes, and 0.4 MTok cache reads. It returned $3.31 per day and $66.10 per month. The break-even script returned 2 hours per month for Max 5x at $100 and a $50 hourly value. Unknown models and negative values exit with errors. These are calculation checks, not billing proof. Confirm subscription charges in Claude Billing and API charges in Claude Console.