Optimizing LLM Costs with Tiered Model Usage

Stop Burning Cash on the Best Model

i was burning money running every coding task through opus. stopped doing that.

now each step goes to cheapest model that can handle it
→ haiku: config, docs, changelog edits
→ sonnet: the script and its tests
→ opus: orchestrate, verify, handle the "risky"

total cost dropped ~80%. the top model just supervised.

wrapped the whole flow into reusable prompt skill. one command triggers it. open sourced here: SKILL.md

switching models for tasks is "hard" and never be easy. they want you to spend tokens.

steal it.

This article was originally published on https://craftengineer.com/. It was written by a human and polished using grammar tools for clarity.
--

Follow me on X (Formally, Twitter). Or read my stories on engineering management, and how to be a better engineering leader on Vibe Manager Blog.