Plan sticker vs API · checked November 2, 2026
Claude Max vs API: Break-Even for Claude Code
Claude Max 5x and Max 20x are plan stickers, not token allowances. Break-even means comparing the API-equivalent cost of your Claude Code tokens with those stickers. Anthropic does not publish a token quota.
Claude Max vs API break-even
Run the Claude Code cost calculator on a real stretch of work. Use Enter tokens, or paste a JSONL log. The total is an API-equivalent estimate at the public list rates on the pricing table. Compare that total with the monthly sticker. If the estimate is under the sticker, the dollar math is under the plan price. That is the whole break-even. It is not remaining included usage, and it is not a promise the workload fits inside the plan.
- Anthropic does not publish token quotas for Pro, Max, or Ultra.
- A five-hour or weekly rate limit can stop Claude Code before the API-equivalent total reaches the sticker.
- Cache reads are most of the tokens in a Claude Code log. Treating every token as uncached input overstates the spend.
| Plan | Monthly sticker | Rough daily line |
|---|---|---|
| Claude Pro | $20.00 | $0.67 |
| Claude Max | $100.00 | $3.33 |
| Claude Ultra | $200.00 | $6.67 |
The rough daily line divides the monthly sticker by 30 and rounds to the cent. Pro is about $0.67 a day, Max is about $3.33 a day, and the $200.00 sticker is about $6.67 a day. That line is a way to glance at one day. It is not how Anthropic bills, and it is not a daily token allowance. Claude Pro annual billing is $200 up front (about $17 a month). Comparisons here use the $20.00 monthly sticker.
Max 5× versus pay-as-you-go API
Claude Max is the Max 5× tier on this card. $100 a month. 5× Pro usage every five hours and every week, plus Claude Code. 5× is a usage-window multiple, not 5× the API dollars. The included $100 of API credits cannot be used in Claude Code.
Pay-as-you-go API cost is the list-rate total for uncached input, output, cache reads, and cache writes. A month of logs under $100.00 is under the Max sticker on this comparison. You can still hit the five-hour window or the weekly window first. 5× is a usage-window multiple. It is not five times the API dollars.
Max 20× and the $200 sticker
Searches for Max 20× are about the $200.00 plan. On this card that plan is Claude Ultra. $200 a month. 10× Pro weekly usage. Verified accounts have no five-hour limit. Replaces the old Max 20× tier. 10× weekly is not a $200 Claude Code credit. Bundled API credits cannot be used in Claude Code.
Use $200.00 when you ask whether a month of API-equivalent Claude Code spend is above or below that sticker. The rough daily line, $6.67, is $200.00 divided by 30. It is not a published daily quota, and it is not 20× the API price. The older Max 20× name is a usage-window label from before Ultra replaced that tier. This page does not treat it as a dollar credit.
One heavy day against those daily lines
The figure below is the same heavy Sonnet 5.5 illustration used on the homepage. It is $8.97 of API-equivalent usage. That single day is above the $0.67 rough daily line for Claude Pro, above the $3.33 rough daily line for Claude Max, and above the $6.67 rough daily line for Claude Ultra. Light and medium mixes are on how much Claude Code costs per day. Your log is the number that matters.
| Token type | Tokens | Rate | Cost |
|---|---|---|---|
| Uncached input | 85,000 | $2.00 / MTok | $0.17 |
| Cache reads | 18,000,000 | $0.10 / MTok | $1.80 |
| 1-hour cache writes | 1,200,000 | $4.00 / MTok | $4.80 |
| Output | 220,000 | $10.00 / MTok | $2.20 |
30 copies of this same day are $269.10. That is the illustration repeated, not a forecast, and not a month you can necessarily run inside a rate limit. Sonnet 5.5 has no long-context surcharge on the published card, so the total scales in a straight line. The calculator's plan comparison for that repeated total:
- $269.10 is 13.45× the $20.00 Claude Pro sticker. No published token quota. Pro has a rolling five-hour limit and a weekly limit. This $20 is the monthly sticker, not a credit balance.
- $269.10 is 2.69× the $100.00 Claude Max sticker. 5× is a usage-window multiple, not 5× the API dollars. The included $100 of API credits cannot be used in Claude Code.
- $269.10 is 1.35× the $200.00 Claude Ultra sticker. 10× weekly is not a $200 Claude Code credit. Bundled API credits cannot be used in Claude Code.
Rate limits can stop you before the dollar math wins
$20 a month, or $200 billed yearly. Includes Claude Code in the same usage pool as chat. No published token quota. Pro has a rolling five-hour limit and a weekly limit. This $20 is the monthly sticker, not a credit balance. Max multiplies the five-hour and weekly windows. Ultra changes the weekly multiple and, for verified accounts, drops the five-hour limit. None of those windows is published as a token count.
A limit can end the session while the API-equivalent total is still under the sticker. The percentage next to a plan in the calculator is API-equivalent cost divided by the monthly sticker. It is not remaining included usage. Max and Ultra API credits stay out of the subtraction, because Anthropic says those credits cannot be used in Claude Code.
Cache reads dominate the token mix
In this heavy illustration, cache reads are 92% of the tokens and $1.80 of the $8.97 total. The 1-hour cache write is $4.80. Pricing every token at the uncached input rate produces $39.01 instead of $8.97. That overstates the spend. Cache reads on Claude Sonnet 5.5 use their own published rate, not the input rate.
The how to estimate page shows where those fields sit in a JSONL log. The FAQ answers why input_tokens looks small.
Estimate your own Claude Code break-even
Open the Claude Code cost calculator. Enter tokens for a day you already counted, or choose Paste or upload a log and drop a JSONL file from ~/.claude/projects/. The file stays in the browser. Nothing is uploaded, and no API key is required. Read the total next to Pro, Max, and Ultra. Then decide whether that API-equivalent figure is above or below the sticker you pay. The limits can still stop you first.