The Skill Tree That Can Cut Your LLM API Bill by 70%

The first thing I thought when starting an AI side project was “how does the cost structure even work?” I’m a developer building a Korean fortune-telling app (saju — four-pillar astrology) and a real estate analysis service as side projects. Both will use LLM APIs. Before writing a single line of service code, I wanted to understand the cost structure. So I dug in. The difference between a pro and an amateur isn’t writing better prompts. It’s engineering the cost structure. Token economics — the foundation of everything Every LLM API charges by tokens. One token is roughly 0.75 English words, or 1-2 Korean characters. Here’s the thing that matters most. Output tokens cost 3-5x more than input tokens. When you say “explain in detail,” the model dumps a wall of text. That’s all output tokens, and that’s all money. If you don’t set max_tokens , the model will write as much as it can. Telling it “answer in under 20 words” is literally a cost optimization technique. Say your fortune service