What it costs

Build-along · ~15 min at your own keyboard

A job that runs on its own also spends on its own. On a flat plan you pay the same each month, but a job that uses an API key, or a tool like Zapier or n8n that holds one, is billed for every run. Most small jobs cost cents a month. The danger is not the normal run. It is the run that loops, retries all night, or reads a huge file by mistake. So do two things before you trust a job: work out what a normal month costs, and set a cap that stops a bad month. Then choose the model on purpose. A cheap model is fine for sorting, pulling out fields and drafts you check. A strong model earns its price when a customer reads the result or it has to make a judgment.

Ladder · Level 4The long version: Automation ↗ · Practice: Route by cost: cheap model, hard model

Your starter: Cost sheet

Template

Make it yours: Fill every line from one real run of your own job. Show the sums, set the cap on the billing page, then write which model you picked and why.

MY JOB:
Runs per week:          x 4.3 = runs per month:
One real run: words in:          words out:
Tokens (words / 0.75): in:          out:

CHEAP MODEL:
Price per million tokens: in $        out $
One run: (tokens in x $in + tokens out x $out) / 1,000,000 = $
Per month: $

STRONG MODEL:
Price per million tokens: in $        out $
One run: (tokens in x $in + tokens out x $out) / 1,000,000 = $
Per month: $

CAP I SET:
Amount per month:          Where:          Date set:

MODEL I PICKED, AND WHY:

The coach checks that what you paste back is your own work. The starter pasted back unchanged won't pass.

Do this

0/6

Checkpoints

▸You should now see runs per month, words in and words out for one real run.
▸You should now see the monthly cost of your job on a cheap model and on a strong model, with the sums shown.
▸You should now see a spending cap that is set, with the amount, where it lives and the date, and the model you picked.

The three ways this goes wrong

You wrote 'it's cheap' and no number. — Do the sum. Cheap is usually true, but you need the number to set a cap that means something, and to notice when a month comes in at ten times normal.
The cap is a plan: 'I'll set a limit later'. — Open the billing page now and set it. A cap you meant to set does not stop a loop at 2 a.m.
Every job uses the strongest model. — Run last week's inputs through the cheap model and compare. If you would send the cheap output as it is, switch. Keep the strong model for what customers read.

Dated facts in this lesson: Claude Haiku 4.5 costs $1 per million input tokens and $5 per million output tokens on the Claude API (researched Oct 5, 2026, not yet confirmed by a person: check the source before you rely on it); Claude Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens on the Claude API (researched Oct 5, 2026, not yet confirmed by a person: check the source before you rely on it); OpenAI gpt-6-luna costs $0.10 per million input tokens and $0.50 per million output tokens on the OpenAI API (researched Oct 5, 2026, not yet confirmed by a person: check the source before you rely on it); OpenAI gpt-6.1-sol costs $2 per million input tokens and $10 per million output tokens on the OpenAI API (researched Oct 5, 2026, not yet confirmed by a person: check the source before you rely on it); Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens on the Gemini API through 31 Dec 2026, then $1.50 and $7.50 from 1 Jan 2027 (researched Oct 5, 2026, not yet confirmed by a person: check the source before you rely on it); A token is roughly 0.75 of an English word. Claude 4.7 and later models count about 30% more tokens for the same text (researched Oct 5, 2026, not yet confirmed by a person: check the source before you rely on it); On the OpenAI API you set a monthly spend limit under Organization limits or Project settings, then Limits, then Spend. With 'Enforce a hard limit' on, requests fail once the limit is reached (researched Oct 5, 2026, not yet confirmed by a person: check the source before you rely on it); On the Claude API you set a monthly spend limit in the Claude Console under Settings, then Billing, then Spend limits. Requests stop when it is reached. Workspaces can have lower limits (researched Oct 5, 2026, not yet confirmed by a person: check the source before you rely on it); Google AI Studio has a monthly project spend cap for the Gemini API. Enforcement can lag by about ten minutes (researched Oct 5, 2026, not yet confirmed by a person: check the source before you rely on it); Microsoft 365 Copilot Chat pay-as-you-go has a budget in the Microsoft 365 admin center that sends alerts. It does not stop billing when the budget is reached (researched Oct 5, 2026, not yet confirmed by a person: check the source before you rely on it).

Share what happened

Paste your cost sheet: runs per month, words and tokens in and out for one real run, the monthly cost on a cheap model and a strong model with the sums shown, the cap you set (amount, where, date), and the model you picked with one line on why.

Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].

🔒 Don't paste confidential, personal-health, or payment info.