Eric Siu explains how to cut AI costs by using expensive frontier models for only 15% of tasks and cheaper models for the remaining 85%. He uses routing tools like OpenRouter to automatically direct queries to the right model based on complexity.
Here's how we save on AI right now. So the way we save on AI is 15% is Ferrari costs, okay? So you have Ferrari models, which is like the frontier models, and then the other 85% is Honda Accord budget, okay? So that's where you like, that's a daily beater that you use to go get your groceries or whatever. So most of the time your people are asking like normal questions, like basic queries, like going back and forth. Maybe sometimes I might ask some stupid questions. That goes in 85%. The 15%, the Ferrari budget, that goes into you actually having the model think for you and code for you. And that's how our infrastructure is like, the cost is breaking down right now. And you might use tools like OpenRouter to route to the right model, and that'll save you more on cost. They'll charge you like 5% or something. That's pretty good.
