Input vs Output Tokens: Which One Is Costing You More?
Almost every model charges more — often three to five times more — for the tokens it generates than for the tokens you send. Knowing which side dominates your workload tells you exactly which lever to pull.
Why output is the expensive side
Generating a token takes more compute than reading one, so providers price output higher. For anything that writes long answers — drafting, summaries, code generation — the length of the reply matters more to your bill than the length of your prompt.
Input-bound vs output-bound workloads
Chatbots that resend history and RAG apps that stuff context are input-bound: trim what you send. Content generators are output-bound: cap reply length and stream. Classify your workload first, then optimize the side that actually costs money.
Calculators for this topic
Frequently asked questions
Why is output more expensive than input?+
Producing tokens is more compute-intensive than reading them, so providers charge a premium — commonly several times the input rate.
How do I cut output cost?+
Ask for shorter answers, set max output tokens, and avoid making reasoning models over-explain.