Sonith ManakkatThere are two quite different ways to pay for Claude, and most confusion comes from mixing them up. This article explains both, shows the prices Anthropic lists today, and ends with a worked example. Each section has the plain picture first and an Expert view with the precise details.
Two ways to pay
Plans (people and teams)
- You chat in the Claude apps
- A monthly price, with usage limits
- Limits reset in five-hour sessions, plus weekly limits on paid plans
The API (software)
- Developers build Claude into their own software
- You pay for the text processed, counted in tokens
- No monthly fee for the model itself
Plans for people
Anthropic's pricing page lists these consumer plans. The page shows "$" without naming a currency and says prices exclude tax.
| Plan | Price as shown | What the page says |
|---|---|---|
| Free | "$0" | Chat on web, desktop and mobile, web search, file creation, code running and memory. Covers everyday questions. |
| Pro | "$17" a month billed annually ("$200 billed up front"), or "$20" billed monthly | At least five times the usage of Free per five-hour session, plus Claude Code, Projects and more models. |
| Max | "From $100" a month | A choice of 5x or 20x the usage of Pro per five-hour session, higher output limits, priority access at busy times. |
Expert viewReading usage limits
Anthropic says all plans have usage limits, and that paid plans add weekly limits. Pro, Max and Team can buy extra usage credits and keep working. Usage depends on the model and the task, which is why long agentic coding sessions use a plan up faster than short chats. The exact limits are not published as fixed numbers on the pricing page; it describes them relatively (for example "at least 5x more usage per 5-hour session than Free"), so budget by observation and read the support articles for your plan.
Plans for teams and organisations
| Plan | Price as shown | Notes |
|---|---|---|
| Team, Standard seat | "$20" per seat a month annually, or "$25" monthly | For teams of 2 to 150, with central billing and single sign-on. |
| Team, Premium seat | "$100" per seat a month annually, or "$125" monthly | Five times the usage of Standard seats. |
| Enterprise | "US$20" per seat a month, billed annually, 20 seat minimum | Extra controls such as audit logs and custom data retention. The page says usage is billed at API rates. |
The API: paying per token
A token is a small piece of text (part 5 explains tokens). Anthropic prices per million tokens, written "MTok", in US dollars on the API pricing page.
| Model | Input, per million tokens | Output, per million tokens |
|---|---|---|
| Claude Haiku 5.5 | $0.10 (prompts up to 100,000 tokens), $0.50 above that | $0.50, or $2.50 above that |
| Claude Sonnet 5.5 | $2 | $10 |
| Claude Opus 5.5 | $4 | $20 |
| Claude Fable 5.1 | $10 | $50 |
Expert viewDetails around the base price
Anthropic's table lists the same base prices as above, plus cache write and cache hit prices. A few points it states: long-context requests on most current models are billed at standard rates, with Claude Haiku 5.5 as the exception (prompts over 100,000 tokens cost more, and the count includes cache reads and writes). Opus 5.5 has a fast mode at higher prices ($8 input and $40 output per MTok, listed as of this check). Specifying US-only inference on 4.6 and later models carries a 1.1x multiplier on all token prices. Models from Claude Opus 4.7 onward use a newer tokenizer that produces about 30 percent more tokens for the same text than earlier models, so always count with the model you plan to use.
- 1Are you building software that calls Claude?Yes The API, paid by tokens processed, not a chat plan.No keep going
- 2Do you need audit logs or custom data retention?Yes The Enterprise plan lists those extra controls.No keep going
- 3A team of 2 to 150 wanting central billing and single sign-on?Yes The Team plans.No keep going
- 4Do you use it heavily and need 5x or 20x the usage of Pro?Yes Max.No keep going
- 5Did you hit Free limits, or want Claude Code, Projects and more models?Yes Pro.No keep going
- 6OtherwiseFree covers everyday questions. Start there.
Takeaway. Match the plan to who is using Claude and how heavily.
Why input and output cost different amounts
Input tokens are everything you send: your message, the earlier conversation, attached files, instructions. Output tokens are what Claude writes, including its reasoning before answering. Output costs more per token. In the list above, output is five times the input price for each model.
- 5xoutput price versus input price, on all four models
- 50%off input and output on the Batch API
- 0.1xor less: price of a cache read versus normal input
- Chatting to Hit the limit Usage depends on the model and the task. Long agentic coding sessions use a plan up faster.
- Hit the limit to Carry on Limits reset in five-hour sessions, and after the reset you can carry on. Paid plans also have weekly limits.
- Hit the limit to Extra credits Pro, Max and Team can buy extra usage credits.
Takeaway. Limits are relative and reset in windows, so budget by watching your own use.
Two discounts worth knowing
Prompt caching
- Store repeated text so re-reading it costs far less
- A cache lasts 5 minutes or 1 hour, depending on what you choose
Batch processing
- Send many requests to be processed later, not instantly
- 50 percent off both input and output tokens
Expert viewCache and batch arithmetic
Anthropic's pricing page lists cache multipliers relative to the base input price: a 5-minute cache write costs 1.25x, a 1-hour write costs 2x, and a cache read costs 0.1x in general (0.05x on Opus 5.5 and Sonnet 5.5, 0.025x on Fable 5.1). It says the 5-minute cache pays for itself after a single reuse and the 1-hour cache after two. Batch and caching discounts can be combined, and the multipliers stack with other modifiers such as data residency. The Batch API is asynchronous, so use it for work that does not need an immediate answer; fast mode is not available with it.
A worked example (illustrative arithmetic)
Suppose a request sends 10,000 input tokens and gets 1,000 output tokens back, on Claude Sonnet 5.5, using the prices above.
- Input10,000 tokens is 0.01 MTok, at $2 per MTok is $0.02
- Output1,000 tokens is 0.001 MTok, at $10 per MTok is $0.01
- Total$0.03 for that request, or about $30 for a thousand of them
The same request through the Batch API would cost half as much. This is just arithmetic to show the shape of the cost. Part 5 includes a small calculator where you enter your own rates.
Which should I choose?
- Just chat, write and learnStart on Free. Move to Pro if you hit limits.
- Use it all dayLook at Pro or Max, and read how usage limits work.
- A teamThe Team plans give central billing and shared controls.
- Building softwareThe API, with the right-sized model and caching.
Plans are for people using Claude. The API is for software using Claude.
Frequently asked questions
Is there a free way to try the API?
Anthropic's pricing page says new users receive a small amount of free credits to test the API.
Do the same words cost the same on every model?
The price per token differs by model, and newer tokenisers can produce about 30 percent more tokens for the same text. Count with the model you plan to use.
Will these prices stay the same?
Not necessarily. Anthropic's page says plans and prices may change.