2 matching entries across Articles, TIL and Projects.
Long-form writing tagged Cost Engineering.
Prompt caching bills writes at a premium and reads at a discount, so a cache that never reads back is strictly more expensive than not caching at all. How I found one deployment doing exactly that, and what made it visible.
Short notes and fixes tagged Cost Engineering.
Through the same Chat Completions endpoint, GPT models counted their hidden reasoning inside completion tokens, and Grok reported it on top. Adding up completion tokens across models adds up different things.
