Prompt engineering is the practice of designing instructions, examples and context so a model produces the right output reliably. Prompt efficiency asks what that output costs. The two meet in four places:
- Stable context first. Put the system prompt, policies and reference material that do not change at the start, so they can be cached and billed at a fraction of the input rate.
- Only the context that helps. Every retrieved passage and every turn of conversation history is billed on each call. Trim what does not improve answers.
- Shorter outputs where they will do. Output tokens cost several times input tokens, and reasoning models add hidden output. Ask for the length and format the task needs.
- The cheapest model that clears the floor. A well-designed prompt often lets a smaller model do work that a vague prompt would need a frontier model for.
The test is always cost per successful outcome, not tokens alone. A prompt that saves tokens but lowers the pass rate moves cost into review and rework. See Token Economics for how each kind of token is billed.