Concepts

Input vs output tokens — a simple guide

Why output tokens often dominate the bill, and how to estimate both sides before you call an API.

Published 2026-07-31

Input tokens

Everything the model reads: system prompts, history, retrieved context, and the user message. Longer RAG contexts raise input spend quickly.

Output tokens

Everything the model writes. Output is often priced higher per million tokens than input, so chatty answers can cost more than large prompts.

How to estimate both

Paste a prompt in the homepage calculator, pick an output preset (classification through long generation), then compare providers side by side.

FAQ

Is output always more expensive than input?

Often yes on a per-token basis, but total spend depends on volume. A huge cached prompt with a short answer can still be input-heavy.