CH 14 · Cost, Latency, and Model Tiering · 11/14
Read as one pageRight-size output
Output tokens are expensive and slow. Don't ask for more than needed: cap output length for bounded tasks (a title needs a handful of tokens, not a paragraph), and instruct the model to be concise where verbosity adds no value. For structured extraction, the format constraints (Chapter 6) also keep output tight.
Finished this idea?
Your learning progress stays in this browser.