Claude subscription vs API: Pro, Max or Ultra, or per token?

The Claude API charges you for every token you use, while a Pro, Max or Ultra plan charges a fixed monthly price in exchange for usage limits. If your use is occasional or you are building your own app, the API is usually cheaper. If you work in Claude or Claude Code for hours every day, a plan usually is.

For the general subscription-vs-API comparison across all vendors, read AI subscription vs API.

What is the difference between a Claude plan and the API?#

At Anthropic, the plans and the API are separate products, each with its own account and its own billing. A plan opens the Claude apps themselves. The API opens the models to code you write.

How much do Claude Pro, Max and Ultra cost?#

These are the individual plan prices as Anthropic publishes them, with the source and the date each one was checked:

Ultra is the plan Anthropic used to call Max 20x. Anthropic describes the Max and Ultra limits as multiples of Pro usage and does not publish a fixed token allowance per plan. So you can't turn a plan price into an exact price per token. The practical comparison is to estimate your monthly usage, then work out what it would cost on the API.

How much does the Claude API cost per million tokens?#

API usage is priced per million tokens, with one price for input (what you send the model) and a higher one for output (what the model writes). These are the newest Claude models on our API, and the full list is on the models page:

And this is the default model's price per million tokens, ours next to the official one:

The default model covers most tasks. The Opus models cost more and suit the hardest ones. Each model has its own page: the default model and Opus.

A worked example: when is the API cheaper?#

Start with light use: a developer who sends two million tokens a month and gets four hundred thousand back on the default model, for occasional questions and small experiments:

At the official price, that month costs less than the Pro plan in the table above. Here the API is cheaper, and you pay only for what you used.

Now take heavy use: daily Claude Code sessions on Opus that send sixty million tokens a month and get six million back:

Before caching, at the official price, that month costs more than three times the official Max price in the table above. That is why people who work in Claude for hours a day usually pick a plan. Anthropic's own average cost per developer for Claude Code is in the Claude Code cost table.

When is a plan the better deal?#

• You work in Claude Code or the Claude apps for hours every day.
• You want a fixed price you know in advance, without watching token counts.
• You don't need the models inside your own app or an automation that runs on its own.

Keep in mind that a plan has usage limits. Once you reach them, you wait for them to reset, buy usage credits from Anthropic, or carry on with an API key.

When is the API the better deal?#

• You are building an app, a service or an automation that calls the model from code.
• Your use comes and goes, with heavy weeks and light ones.
• You use models from more than one provider, such as GPT and Gemini alongside Claude.
• You want a spend limit per project, or a separate key for each person on your team.

A plan and the API through 5abeer#

5abeer is an independent service, not affiliated with Anthropic, and it offers both routes below the official price.

The API: you top up your balance, create a key for Claude models, and each request is deducted by model and token count. You control every key with a spend limit, an expiry date and an IP allow-list. All prices are on Claude API and the API page.

A plan: a monthly plan that runs the official apps, such as Claude Code, through the 5abeer AI app, with a five-hour limit and a weekly limit. Plans are for the supported apps; to use the models from your own code you need an API key. Pick a plan on the pricing page, and read the details in How a plan works.

Your balance and your plan are separate: API key usage never counts against your plan's limits, and a plan only draws on your balance if you turn on Continue from the wallet.

How to decide in three steps#

1. 1

   Estimate your monthly usage#

   Look at a normal month of work. If you run Claude Code on an API key, the Session block of the /usage command estimates what the current session has cost at list price. If you already use the API, your dashboard shows usage per key and per model.

2. 2

   Price it on the API#

   Multiply your millions of input tokens by the input price and your millions of output tokens by the output price, then add the two. The full method is in How cost is calculated.

3. 3

   Compare with the plan price#

   If the result is below the price of the plan you would need, the API is cheaper for you. If it is well above it, and your work happens inside the supported apps, a plan is cheaper.

FAQ#

Does Claude Pro include API usage?#

No. At Anthropic the plans and the API are separate, and the API is paid for in the Claude Console. Anthropic does say that Max and Ultra include monthly API credits, as shown in the table above.

Does Claude Code work with a plan or with an API key?#

Both. You can sign in with a plan account, or run it on an API key that draws on your balance. Key setup is in Claude Code setup, and the two routes are compared for coding tools in The cheapest way to use Claude Code.

Is Opus more expensive than the default model?#

Yes. Opus costs more per million tokens, and the gap shows in the models table above. Start on the default model and move to claude-opus-5.5 or claude-fable-5.1 when the task calls for it.

What happens when my plan's limits run out?#

You wait for the limits to reset, buy usage credits from Anthropic, or carry on with an API key that draws on your balance. On 5abeer you can also turn on Continue from the wallet, so usage beyond your plan's limits is deducted from your balance.

Can I use a plan inside my own app?#

No. A plan is for the supported official apps. To call the models from your own code or an automation, you need an API key. Start with the Quickstart.

How do I lower my API cost?#

Use caching for repeated context, start on the default model before Opus, and cap the output length when a short reply is enough. A spend limit on each key keeps any one project from running past its budget.

Get started#

Compare Claude model prices on Claude API, or plan prices on the pricing page.