AI Cost CalculatorThe formula
Here's the simple version: for each request you pay for the words you send in and the words the AI sends back, each at the model's own price. Multiply that by how many requests you make in a month, add a small safety cushion, and that's your bill. The extra options (retries, tool calls, search, human checking) just add a bit more on top, and caching or batch jobs take a bit off.
Questions
Why does the AI's reply cost more than what I send in?
Writing an answer takes the model more effort than reading your prompt, so most providers charge a few times more for the reply. The upshot: asking for shorter answers usually saves you more than shortening your question.
Can I trust this monthly number?
It's a solid estimate, not a quote. Your real bill moves with how much people use the app, how often calls fail, how long each message is, and which model you pick - so keep the safety cushion on and check your usage once you're live.
How do I make it cheaper?
Use a smaller model for the easy work, ask for shorter answers, reuse repeated text with caching, run non-urgent jobs as batch, and only look up the knowledge you actually need.