Claude API Pricing vs Subscription: Pro, Max and Ultra Compared

Claude Pro costs $20 a month. On the API, $20 a month pays for about 51 messages a day on Claude Sonnet 5.5, if each message sends 4,000 tokens and gets 500 back (our example size, not a measurement). On Opus 5.5 it pays for 25, and on Fable 5.1 for 10.

That answer depends on what a message weighs, what the plans include besides chat, and the API credit that comes with Max and Ultra. The math is below, so you can redo it with your own habits. Which one you buy is your call.

What each Claude plan costs, and what it covers

Anthropic sells three paid plans for individuals, plus the pay-as-you-go API. Plan prices below are US, monthly, web billing; everything was read on Anthropic’s own pages on 2026-11-03:

• Claude Pro, $20. More usage than the free plan, a five-hour limit and a weekly limit. The Pro plan page says plainly that it does not include API usage.
• Claude Max, $100. Five times Pro’s usage every five hours and every week, per the Max plan page, plus $100 of API credit a month.
• Claude Ultra, $200. Ten times Pro’s weekly usage and no five-hour limit for verified accounts. The Ultra plan page says this is the plan formerly called Max 20x. It comes with $200 of API credit a month.
• The API has no monthly fee. You pay per million tokens, input and output priced separately. The pricing page lists Sonnet 5.5 at $2 input and $10 output, Opus 5.5 at $4 and $20, and Fable 5.1 at $10 and $50.

One thing makes this an awkward comparison: the plans don’t state their limits in tokens or dollars. The Pro page says the number of messages you can send varies with message length, attachments, conversation length and the model you use. So nobody outside Anthropic can tell you that Pro equals exactly N dollars of API. What you can work out is what your own use would cost on the API, and compare that with the plan price.

Claude API pricing per message

A message on the API costs (input tokens × input price + output tokens × output price), with prices quoted per million tokens. The input side is bigger than people expect. Anthropic’s Messages API guide (checked 2026-11) says the API is stateless, so every request carries the full conversation so far. Your tenth message pays again for the nine before it.

Take an example message, a few turns into a chat, that sends 4,000 input tokens (your new question plus the earlier turns) and gets back a 500-token reply. By our tokens-per-word measurement, 500 tokens on Claude Opus 5.5 hold about 290 English words (an estimate). At that size, $20 buys 51 messages a day on Sonnet 5.5 for a 30-day month:

Read the table as a break-even line. If you send fewer messages than the number in a cell, that amount of API usage costs less than the plan. If you send more, the plan costs less, if its limits allow that many, which Anthropic doesn’t say. Max and Ultra also cap Fable separately each week. The model matters as much as the plan: Fable 5.1 costs five times what Sonnet 5.5 does per message, so Pro’s price covers only 10 Fable messages a day.

What a daily chat habit costs on the API

Now turn it around. Say you send 40 messages a day of the same size, every day of the month. On Sonnet 5.5 the API bill comes to $15.60, under Pro’s $20. Move the same habit to Opus 5.5 and it doubles to $31.20; on Fable 5.1 it reaches $78.

Long conversations push every bar up, because each new turn adds the previous reply to the input of the next request. A fresh conversation resends less history, so each message costs less, and Anthropic’s Pro page says conversation length also counts against plan limits.

Max and Ultra include API credit, which changes the math

Anthropic’s API credits page (checked 2026-11-03) says Max includes $100 and Ultra $200 of Claude API credit each month. You claim it on claude.ai by linking a Claude Console organization, and new subscribers can claim it after 7 days on the plan. Pro and Free plans aren’t eligible.

• Unused credit expires at the end of each billing cycle. Nothing rolls over.
• It covers the Claude API, Managed Agents, the Agent SDK and the Console playground. It does not cover Claude Code or extra usage in the Claude apps.
• It works only on the Claude API in the Claude Console, not on Claude Platform on AWS, Amazon Bedrock, Google Cloud or Microsoft Foundry. Credits also don’t change your usage limits in Claude or Claude Code.

So if you already spend $100 or more a month on the Claude API in the Console (Claude Code doesn’t count), Max costs nothing extra for the chat app on top, as long as you use the credit every month. If your API spend is a few dollars, most of the credit expires unused.

What the price comparison leaves out

The plans include things the API doesn’t: the Claude apps, Claude Code access and longer tasks that keep running in the background, all in one subscription. The API gives you things the plans don’t: your own system prompt, your own code, volume up to your tier’s rate limits, and discounts. On the pricing page, a cache hit on Sonnet 5.5 costs $0.10 per million tokens against $2 for fresh input, so a script that reuses a long prompt can cost much less than the table shows. Cutting LLM API costs covers caching and the other levers.

There is also a cheaper model the table leaves out. The same pricing page lists Claude Haiku 5.5 at $0.10 input and $0.50 output per million tokens for prompts up to 100,000 tokens, a twentieth of Sonnet 5.5’s rates. If a small model handles your questions well, the API costs far less. Whether it does is something only your own prompts can tell you.

Work out your own break-even

Open the LLM Cost Calculator, switch to Paste a prompt and paste a real message from late in one of your conversations, with the earlier turns included, since that’s what the API would be sent. Set the output to the reply length you usually get and the requests per day to how often you write to Claude. Pick a Claude model and the calculator lists Pro, Max and Ultra next to your monthly estimate, and says whether Max’s or Ultra’s credit would cover it. Nothing you paste leaves your browser.

Our Claude token counts are estimates, because Anthropic doesn’t publish a tokenizer that runs in the browser. Counting Claude tokens explains how close they get and how to ask Anthropic for the exact number when it matters.