PromptPinchStart with tokens
AI providers bill by tokens, which are small pieces of text. Every request has two parts: the text you send in and the text the model writes back. Providers list a price per million tokens for each model.
Input and output are priced differently
Most providers charge more for output tokens than for input tokens. A short question that produces a long answer can cost more than a long question that produces a short one. Some newer models also count their internal reasoning as output, which can raise the cost of a single answer.
Models are priced differently
The same request can cost many times more on a large model than on a small one. When a team picks one model for everything, simple tasks end up priced like hard ones.
Your bill does not say which app spent it
Provider invoices usually show totals by model, not by app. If three apps share one account, you cannot tell which one is growing. Giving each app its own key, or reporting by app, is the first step to controlling spend.
What to look at each month
- Total spend and the change from last month
- Spend by model
- Spend by app or team
- Cost per request or per customer conversation
If you can answer these four, you know where your AI budget goes. If you cannot, that is the place to start.