Understanding costs
Every time you run a helper, Margaret sends your input and the helper's instructions to Claude, and Claude writes a response. Anthropic charges for this based on the amount of text involved — both going in and coming out. These units of text are called tokens.
You don't need to think about tokens day-to-day. Margaret translates them into approximate word counts wherever it shows them, and shows you a cost estimate after every run.
What is a token?
A token is roughly three-quarters of a word. "Margaret" is one token. "I need help with my weekly report" is about eight tokens.
The reason it matters: Anthropic prices by the thousand tokens, separately for input (what you send in) and output (what Claude writes back). Output tokens cost more than input tokens.
What does a typical run cost?
It depends on the model, the length of your input, and how much the helper writes back. As a rough guide:
| Task type | Model | Approximate cost |
|---|---|---|
| Short formatting task (1–2 paragraphs in, 1 page out) | Haiku | $0.001 – $0.003 |
| Standard analysis (1 page in, 2 pages out) | Sonnet | $0.005 – $0.015 |
| Long document review (10 pages in, detailed report out) | Sonnet | $0.03 – $0.08 |
| Complex strategic brief (detailed input, long output) | Opus | $0.05 – $0.20 |
Most runs for everyday business tasks cost between one and five cents. Running a helper ten times a week typically costs less than a coffee per month.
Prompt caching
If you run the same helper multiple times, Margaret automatically benefits from Anthropic's prompt caching. The helper's instructions only need to be processed once — subsequent runs that use the same instructions are charged at a fraction of the normal rate.
Margaret shows "some input cached" in the run summary when this kicks in. You don't need to do anything to enable it.
Keeping an eye on your spend
In Margaret: The dashboard shows your total spend, a breakdown by helper, and the cost of every recent run. You can always see exactly where your money went.
In Anthropic's console: Visit console.anthropic.com → Usage to see your full billing history and current balance.
Tips for keeping costs low
Use the right model for the task. Haiku handles simple, structured tasks well and costs a fraction of Sonnet. Don't use Opus for tasks that don't need it. See Choosing a model for guidance.
Keep inputs focused. If your helper only needs the summary of a document, don't paste the entire document. Longer inputs cost more.
Check your output format. If a helper is writing pages when you only need a few bullet points, tighten the Output Format section in its instructions.
Use prompt caching. It happens automatically — just make sure you're running helpers through Margaret rather than pasting their instructions into Claude.ai each time.
I didn't expect that charge — what happened?
The most common reason for a higher-than-expected cost is a very long input. Pasting a large document into the paste field, or picking a large file, increases the input token count significantly. Check the run details in the dashboard — it shows input tokens, output tokens, and cost per run.
If something looks wrong with your Anthropic billing, contact Anthropic support directly at support.anthropic.com.