Claude Max vs API list prices: what a coding task costs per token

You cannot convert a Claude Max plan into an exact cost per task, because the published limits are relative (5x or 20x Pro usage per five hour session), not token counts. You can price a task at API list rates, then compare it with the flat monthly fee. All prices below were fetched on 18 October 2026.

What do the plans cost?

From Anthropic's pricing page, fetched 18 October 2026:

• Pro: $17 per month with annual billing ($200 billed up front), or $20 if billed monthly.
• Max: from $100 per month, with 5x and 20x options. The page does not list a separate price for each tier. Max 5x and 20x give 5x or 20x more usage than Pro, with higher output limits and priority access at busy times.
• API credits on Max: $100 per month on Max 5x and $200 per month on Max 20x, for building on the Claude Platform.
• Limits: every plan has usage limits that reset on a rolling five hour session window, and paid plans add weekly limits. Usage is shared across web, desktop, mobile and Claude Code. When you hit a limit you can wait, upgrade, or turn on usage credits.

Claude Code is included on paid plans and shares the same limits as chat.

What are the API list prices?

From Anthropic's API pricing documentation, fetched 18 October 2026, per million tokens (MTok):

• Claude Sonnet 5.5: input $2, output $10
• Claude Opus 5.5: input $4, output $20
• Claude Fable 5.1: input $10, output $50

The page also lists cache hit prices (for Sonnet 5.5, $0.10 per MTok) and says the Batch API gives a 50 percent discount on both input and output tokens for asynchronous work.

What does one task cost at list price?

This is an illustration, not a measurement. Suppose an agent task reads 200,000 input tokens and writes 20,000 output tokens. That is 0.2 MTok in and 0.02 MTok out.

• Sonnet 5.5: 0.2 times $2 is $0.40, plus 0.02 times $10 is $0.20. Total $0.60.
• Opus 5.5: $0.80 plus $0.40. Total $1.20.
• Fable 5.1: $2.00 plus $1.00. Total $3.00.

If 160,000 of those input tokens are cache hits on Sonnet 5.5, the input costs 0.04 times $2 ($0.08) plus 0.16 times $0.10 ($0.016), and output stays $0.20, for about $0.30.

Real agent sessions vary a lot, because every step re-reads context. Your own logs will give better numbers than ours.

Can you compare that with a flat fee?

Only loosely. Divide the monthly price by your per task list cost to get a break even count. At $100 per month and $0.60 per task, that is about 166 tasks a month. At $1.20, it is about 83. If you run more than that at API list rates, a flat plan could be cheaper, and fewer than that, the API could be.

The comparison breaks in three places:

1. The plan limits are not published in tokens, so you cannot say how many tasks a plan includes.
2. Limits reset on five hour and weekly windows, so heavy bursts can hit them even if monthly volume looks low.
3. The plan includes things the API price does not, such as the chat apps. The other direction also holds: API usage has no session cap.

We make no claim about what any provider spends to serve a task. Only published prices are used here.

What to do

1. Pull a week of your own usage logs and total input, output and cached tokens per task.
2. Price those tokens at the list rates above for the model you use.
3. Compare the monthly total with the plan fee, and note how often you hit session or weekly limits.
4. Decide per workload. Interactive daily coding and bulk overnight jobs often suit different options, and the Batch API discount applies only to asynchronous work.
5. Recheck prices each quarter, since they change.

Quick check

Sources

• Claude pricing page, fetched 18 October 2026
• Anthropic API pricing documentation, fetched 18 October 2026

Last reviewed: 18 October 2026.