Claude Code starts at Claude Pro, which costs $20 per month. Max 5x costs $100 and Max 20x costs $200, and each includes API credits equal to its price. At API rates, one modeled 50-step coding agent run on Claude Sonnet 5.5 costs 84 cents with caching, so $100 buys about 119 runs a month.
What the Claude plans cost
The free plan does not include Claude Code.
- Claude Free, $0 per month: No Claude Code. No Opus or Fable access.
- Claude Pro, $20 per month: Claude Code included, with more usage than Free. Fable is available only via usage credits.
- Claude Max 5x, $100 per month: 5x more usage than Pro, plus $100 per month in API credits for the Claude Platform. Fable is limited to 50% of weekly limits, and you get priority access at high traffic times.
- Claude Max 20x, $200 per month: 20x more usage than Pro, plus $200 per month in API credits for the Claude Platform. Same Fable limit and priority access as Max 5x.
A plan is a flat monthly price. The API bills per token, so cost depends on the model and the length of the work. To compare the two, I need the API cost of a typical agent run.
What one agent run costs on the API
My 50-step agent task is a modeled 50-step coding agent. It starts with an 8,000-token system prompt. Each step adds 1,500 tokens of context and 300 visible output tokens. Reasoning tokens are billed as output tokens.
| Model | Per run, cached | Per run, no caching | Per 1,000 runs (cached) | Input / cached input per 1M tokens |
|---|---|---|---|---|
| Claude Haiku 5.5 | 6.6¢ | $0.34 | $66.35 | $0.10 / $0.01 |
| Claude Sonnet 5.5 | 84¢ | $6.52 | $841.64 | $2.00 / $0.10 |
| Claude Opus 5.5 | $1.68 | $13.04 | $1,683.28 | $4.00 / $0.20 |
The median of all 326 models on this task is 48¢ per run.
Caching matters most here. An agent resends the same system prompt and history on every step, so most input tokens can be billed at the cached rate. On Sonnet 5.5 that is $0.10 per million tokens instead of $2.00, which is why a run drops from $6.52 to 84¢. To test your own numbers, use the calculator.
Model details: Claude Sonnet 5.5, Claude Opus 5.5, Claude Haiku 5.5.
Break-even: runs per month
The table shows how many cached 50-step agent runs each dollar amount buys at API rates. For Max, I use the included API credits. My data lists no API credit line for Pro, so the Pro row shows what $20 would buy at API rates.
| Plan | Haiku 5.5 runs | Sonnet 5.5 runs | Opus 5.5 runs |
|---|---|---|---|
| Pro ($20) | about 300 | about 24 | about 12 |
| Max 5x ($100) | about 1,500 | about 119 | about 59 |
| Max 20x ($200) | about 3,000 | about 238 | about 119 |
Reading the table:
- Sonnet 5.5: About 119 runs a month is where $100 of API spend equals the Max 5x price.
- Opus 5.5: About 59 runs equals $100, and about 119 runs equals $200.
- Haiku 5.5: Even hundreds of runs cost far less than any plan price.
Note that the Max credits are for the Claude Platform API. They are separate from the plan's usage limits, so the price is not a direct purchase of runs inside Claude Code.
When the API is cheaper, and when a plan is
The API is the better fit when:
- You run only a few agent jobs a month. At about 24 Sonnet runs, API spend reaches Pro's $20.
- You use Haiku 5.5 for most work. At 6.6¢ per run, it sits well below the plan prices.
- You need to control spend per job through your own code and budgets.
A Max plan is the better fit when:
- You run Opus 5.5 or Sonnet 5.5 at high volume. Past roughly 59 Opus runs or 119 Sonnet runs a month, API spend exceeds $100.
- Your sessions are interactive and irregular, and a flat price is easier to manage than per-token bills.
- You want the included API credits plus higher Claude Code usage in one price.
One limit: my data does not state the token or message quota behind each plan's usage limit, so I cannot convert Pro, Max 5x or Max 20x into an exact run count. The table compares plan prices against API rates. Real sessions vary in length, tool use and cache hit rate, so your numbers may land above or below these.
Methodology
The 50-step agent costs are modeled from the scenario above, not measured. Prices and costs are from the TaskCents dataset as of 2026-10-11, in USD. See methodology for how I treat caching, reasoning tokens and modeled tasks, and plans for current plan details.
FAQ
Is Claude Code included in the Pro plan?
Yes. Claude Pro costs $20 per month and includes Claude Code, with more usage than Free. The free plan does not include it.
How much does one agent run cost on the API?
In my modeled 50-step test, a run costs 84¢ on Sonnet 5.5, $1.68 on Opus 5.5 and 6.6¢ on Haiku 5.5, all with caching. Without caching, Sonnet 5.5 costs $6.52.
Is Max 5x cheaper than the API for heavy use?
For heavy use of Sonnet 5.5 or Opus 5.5, usually yes. Max 5x costs $100 and includes $100 in API credits, which covers about 119 Sonnet runs or about 59 Opus runs at cached API rates. Beyond that volume, the same work costs more than $100 on the API.