I spent a chunk of last weekend with about fifteen pricing pages open in different tabs, trying to figure out why a small side project’s API bill quietly tripled in a month. Turns out the answer wasn’t “the model got more expensive” — it was that I picked the wrong model for the job, on the wrong platform, without caching anything. So I went down the rabbit hole properly: every major LLM API, every cloud wrapper, and the budget options nobody talks about until their AWS bill shows up.