Is my Claude plan worth it?

How the estimate works

Your token counts come from ccusage, an open-source command-line tool that reads the usage records Claude Code keeps on your computer and totals them by day or month. Its JSON report lists, for each period, a modelBreakdowns array with inputTokens, outputTokens, cacheCreationTokens and cacheReadTokens per model.

The calculator reads that report and adds up every model across all periods. Newer ccusage versions can also report Codex usage. In that format each period has a combined row with each tool’s rows nested inside it. The calculator counts the combined row once and skips the nested copies. If you paste output that holds several reports at once, it counts only one (monthly first), so nothing is counted twice.

Model names are normalized the same way the Quotchi app does it: claude-opus-4-8-20260101 becomes Opus 4.8, and Bedrock or Vertex suffixes are dropped. Only an exact match gets a price. A model the price table doesn’t list, such as gpt-5-codex or a Claude model released after the table’s date, shows as not priced and adds $0. It never borrows a neighboring model’s price.

The pricing formula

Anthropic prices five kinds of tokens separately, in US dollars per million tokens (MTok). For each model:

value = ( input × input price + output × output price + cache write (5-minute) × 5m write price + cache write (1-hour) × 1h write price + cache read × cache read price ) ÷ 1,000,000

Prompt caching is why Claude Code’s token counts look enormous. Each turn re-sends the conversation so far, and most of it is served from the cache. A cache read costs a fraction of normal input: one tenth on most models, less on some newer ones. A cache write costs more than normal input: 1.25× for the 5-minute lifetime and 2× for the 1-hour lifetime. That split matters, because cache reads are usually most of your tokens but a smaller share of the value.

The calculator holds the full table: 20 Claude models from Haiku 3.5 to Fable 5.1, copied from Anthropic’s pricing page as read on October 9, 2026. Prices are standard global-endpoint rates, without Batch discounts, fast mode or regional surcharges. Claude Haiku 5.5 costs more for prompts over 100,000 tokens. Summed counts can’t show which requests were that long, so Haiku 5.5 is priced at its standard rate and the result says so.

A worked example

The Try sample JSON button loads a made-up month. Here is its Opus 4.8 row, priced with 5-minute cache writes:

input 1.2M × $5.00 = $6.00 output 0.85M × $25.00 = $21.25 cache write 9M × $6.25 = $56.25 cache read 120M × $0.50 = $60.00 Opus 4.8 = $143.50

Sonnet 4.6 adds $28.95 (0.4M input, 0.3M output, 3M writes, 40M reads), and Haiku 4.5 adds $1.15. The sample also contains gpt-5-codex, which is not priced. The total is $173.60 for September 2026. That’s 8.7× the Pro price, 1.7× Max 5x and 0.87× Max 20x. Price the same writes at the 1-hour rate and it rises to $214.40.

Notice that cache reads were 120M of Opus’s 131M tokens but only $60 of its $143.50. Raw token totals overstate value; the formula keeps them honest.

The plan prices we compare against

• Pro: $20 per month billed monthly, or $17 per month with annual billing ($200 up front). claude.com/pricing
• Max 5x: $100 per month. Max 20x: $200 per month. Monthly only. Claude Help Center

These are US web prices before tax; Anthropic notes mobile pricing may differ. Last checked October 9, 2026. When your report covers more or less than a month, the plan bars use a per-month figure: the total divided by the number of months, or scaled from the number of days in a daily report.

What this tool can’t know

• Your remaining quota. Anthropic doesn’t publish Pro or Max allowances in tokens or dollars. Limits are described in relative terms (Max 5x and Max 20x are described as five and twenty times Pro’s usage), with a five-hour session window and a weekly limit. No token count converts into “% left”. How Claude Code limits work.
• Usage on other machines or apps. ccusage only sees Claude Code records on the computer where you run it. Chats on claude.ai, in the desktop app and on other computers aren’t included, though they share the same plan limits.
• The cache-write lifetime. ccusage gives one cache-write number, so you choose how to price it.
• What you would really have paid. On the API you might pick other models, use Batch pricing, or simply work differently. The estimate prices what you did, at list price, on the day the table was read.

Tips for a fair comparison

• Use a whole month. A daily report from a busy week projects a busy month that may never happen.
• Combine machines by running ccusage on each and adding the results, or by entering the totals by hand in Enter tokens.
• Compare against ccusage’s own cost figure, shown under the result when the report includes it. A big gap usually means a different price list or cache-write assumption.
• Look at headroom as well as value. A plan that costs less than your list-price value is good value. Whether you also keep hitting limits is a separate question: check /usage, or when your limit resets.

FAQ

Sources

Read on October 9, 2026. If anything here disagrees with them, they win.

• Anthropic: Pricing (model price table, cache write and read rates)
• Claude: Plans & pricing (Pro)
• Claude Help Center: What is the Max plan? (Max 5x and Max 20x)
• ccusage: JSON output format