How Much Does Claude Code Cost Per Month?

Every article about Claude Code pricing quotes the same headline numbers: $20/month for Claude Pro, $100/month for Max. What those articles don't tell you is that the subscription price is only half the equation. The other half is how many tokens you actually burn — and whether the flat rate is cheaper than what you'd pay at API prices per token. That math is different for every developer, and you cannot do it without measuring your own usage.

What actually drives your monthly cost?

Claude Code charges differ based on three variables: which plan you're on, how many sessions you run each day, and how deep into context each session goes. A developer who opens a 200k-token context for an afternoon of refactoring burns dramatically more than one who runs a handful of short, targeted tasks. Before you can reduce or budget your Claude Code cost, you need your actual weekly totals.

Why weekly totals matter more than daily averages

Usage spikes around deadlines and drops on weekends. A flat daily average hides the weeks when you burned 40M tokens during a sprint and the weeks you barely opened the tool. The bar chart below represents a typical developer month — notice the spike in week 2 when a big feature landed:

In this example the month totals roughly $93 at API rates. If you're on a $100 Max plan, you're close to breaking even. If you're on a $20 Pro plan, the math doesn't work. The point: you cannot know without measuring week 2 specifically.

The cost equation for your month

Whoburnedmore reads the JSON logs Claude Code writes to your home directory and applies per-model token prices to give you the API-equivalent cost. The full monthly cost is the sum across all days and all models you used:

Input tokens and output tokens are priced differently — on Claude's Sonnet models, output tokens cost roughly five times more per token than input tokens. Sessions that generate long code completions cost more per token burned than sessions that mostly read context. Whoburnedmore breaks this down per model so you can see exactly where the dollars went.

Subscription vs API: the breakeven calculation

On a flat-rate Claude Pro or Max plan, the “cost” whoburnedmore reports is what your usage would have cost at API rates — not what you were actually billed. The relevant question is whether the subscription saves money:

How to measure your real monthly spend

Run the tool once to get your 30-day snapshot. It reads the local usage logs Claude Code already maintains — no installation required beyond npm:

$ npx whoburnedmore↳ reading ~/.claude/usage/ …   MONTH OVERVIEW (last 30 days)  model             input tokens    output tokens   est. cost  ─────────────────────────────────────────────────────────────  claude-sonnet-4   312.1M          42.8M           $71.40  claude-haiku-4     18.4M           3.1M            $2.10  ─────────────────────────────────────────────────────────────  total             330.5M          45.9M           $73.50   ℹ  on Claude Max ($100/mo): saving ~$26.50 this month

1. 1

   Run npx whoburnedmore

   No install needed — npx fetches and runs it. You'll see a per-model breakdown for the last 30 days in about 5 seconds.
2. 2

   Note your API-equivalent total

   That total is what you'd pay if you were on the API instead of a subscription. Compare it to your actual plan price.
3. 3

   Check week-over-week trend

   Look at the weekly breakdown. If one sprint week was an outlier that dragged your total up, the average month might actually be cheaper.
4. 4

   Decide if your plan fits your usage

   If your API-equivalent cost is consistently higher than your subscription, the flat plan wins. If it's consistently lower, downgrade or switch to pay-as-you-go.

Plan comparison for different usage levels

The right plan depends entirely on your monthly token volume. The table below gives rough breakeven points using illustrative API-equivalent rates for Sonnet-class models. Actual prices may vary — run npx whoburnedmore to see your real numbers.

What makes costs spike unexpectedly?

Developers are regularly surprised by how much a single long session costs. There are three main culprits, all of which show up in the whoburnedmore breakdown:

Large context windows on every turn

Claude Code re-sends the entire conversation context on each request. A session that reaches 100k tokens of context burns 100k input tokens on every subsequent message, even if your new message is just five words. Long sessions with a full codebase loaded are the biggest driver of unexpected cost.

Repeated file reads across sessions

If you start a fresh session but immediately ask Claude to read the same large files again, you're paying for those input tokens again. Context isn't persisted across sessions by default. Measuring daily totals in whoburnedmore will show you which days you repeated expensive reads.

Using Sonnet for tasks that Haiku handles

Not every task needs the most capable model. Short grep-style lookups, writing boilerplate, or checking syntax can run on Haiku at a fraction of the cost. If your model split shows 100% Sonnet usage, you might be over-spending.

Tracking your monthly Claude Code cost doesn't require a spreadsheet or a third-party dashboard. The logs are already on your machine. Running npx whoburnedmoreonce a week takes under 10 seconds and gives you an accurate picture of where your subscription budget is going — and whether it's going to the right place. 📊