On 18 October 2026, Anthropic's pricing page lists Claude Pro at $20 a month (or $17 a month billed annually), and Max "from $100 a month" with a choice of 5x or 20x the usage of Pro. The Claude API is billed per million tokens instead. If you code with an agent all day, which of the three is the sensible way to pay?
This post uses only Anthropic's own pages, fetched on 18 October 2026. Anthropic does not publish how many tokens a plan allows, so nothing below states a token limit for a subscription. Where a number is mine, it is labelled as a scenario.
What Anthropic publishes
From claude.com/pricing and the Max plan help article:
| Plan | Published price | Published usage wording |
|---|---|---|
| Pro | $20 monthly, or $17 a month on annual billing ($200 up front) | At least 5x more usage per 5-hour session than Free. Claude Code included |
| Max 5x | $100 a month | "five times the Pro plan's per-session usage allowance" |
| Max 20x | $200 a month | "20 times the Pro plan's per-session usage allowance" |
| Team premium seat | $125 monthly, or $100 on annual billing | 5x more usage than a Team standard seat |
Anthropic also states that "every plan has usage limits that reset on a rolling five-hour session window, and paid plans add weekly limits on top." The Max help article adds that Anthropic may impose other caps "at our discretion." There are no message or prompt counts in the article, only multiples of Pro.
What the API charges
Per million tokens (MTok), from platform.claude.com pricing, fetched the same day:
| Model | Input | Output | Cache read | 5 minute cache write |
|---|---|---|---|---|
| Fable 5.1 | $10 | $50 | $0.25 | $12.50 |
| Opus 5.5 | $4 | $20 | $0.20 | $5 |
| Sonnet 5.5 | $2 | $10 | $0.10 | $2.50 |
| Haiku 5.5 (prompts up to 100K tokens) | $0.10 | $0.50 | $0.01 | $0.125 |
Batch processing is listed at a 50% discount, but a coding agent is interactive, so batch rarely applies to it.
A scenario you can change
Coding agents re-read the conversation on every step, so most input tokens are cache reads. Suppose one working day, for one developer, looks like this. These inputs are my assumptions, not Anthropic data:
- 3,000,000 cache read tokens
- 300,000 fresh input tokens
- 50,000 cache write tokens
- 100,000 output tokens
- 22 working days a month
prices = { # USD per MTok: fresh in, cache read, cache write, out
"Sonnet 5.5": (2.00, 0.10, 2.50, 10),
"Opus 5.5": (4.00, 0.20, 5.00, 20),
"Fable 5.1": (10.00, 0.25, 12.50, 50),
}
read, fresh, write, out, days = 3.0, 0.3, 0.05, 0.1, 22 # MTok per day
for m, (i, r, w, o) in prices.items():
day = fresh*i + read*r + write*w + out*o
print(f"{m}: ${day:.2f} a day, ${day*days:.2f} a month")
With these inputs the API list price comes to about $44.55 a month on Sonnet 5.5, $89.10 on Opus 5.5 and $206.25 on Fable 5.1. Double the tokens and the numbers double. The point is the method, not the result: a light user may sit far below a plan price, and a heavy agent user may sit far above it.
How to choose
- Pro suits people who code in bursts. Start here and watch whether you hit the five-hour limit.
- Max suits people who hit Pro limits most days and want a fixed monthly bill with a predictable ceiling.
- API suits automation, CI, shared services and anyone who needs to see every token. You pay for exactly what you send, and you can pick a cheaper model per task.
One more difference is data terms. Anthropic's pricing page lists model training as "Opt-out" for Free, Pro and Max, and "No model training on your content by default" for Team. Check the current wording before you put company code through a personal plan.
If you use Safeguard, you can connect it to Claude through MCP, where tool calls are unlimited, or bring your own model. That keeps the model bill in your hands, whichever way you pay for it.
Quick check
Does Anthropic publish a token limit for Max?
No. The Max help article states limits only as multiples of the Pro per-session allowance, plus weekly limits, and gives no message or token counts.
