# Cost, Latency and Prompt Length — Prompt Engineering

Source: https://www.geekswithgeeks.com/en/prompt-engineering/v-cost

> Trim prompts and choose models to keep systems affordable and fast.

## Every token is paid for, every call

Providers charge per input and output token, so a long fixed prefix is paid for on **every call**. Reduce cost and delay by: cutting redundant instructions and unneeded examples, shortening retrieved context, capping **max output tokens**, using a **smaller model** for easy tasks, **caching** repeated prefixes (prompt caching) where available, and batching non-urgent work. Always re-run the test set after shortening a prompt to confirm quality held. The numbers below use example prices, not any provider's real price list.

## What a shorter prompt saves, run

I ran this plain-Python (standard library only) example. At the example prices, cutting the prompt from 1,800 to 600 tokens saves $360 per 100,000 calls (from $840 to $480).

```python
def cost(prompt_tokens, output_tokens, price_in=3.0, price_out=15.0, calls=1):
    return calls * (prompt_tokens * price_in + output_tokens * price_out) / 1e6

long_p, short_p = 1800, 600
print("long prompt per 100k calls: $", round(cost(long_p, 200, calls=100_000), 2))
print("short prompt per 100k calls: $", round(cost(short_p, 200, calls=100_000), 2))
print("saving: $", round(cost(long_p, 200, calls=100_000) - cost(short_p, 200, calls=100_000), 2))

```

Output:

```
long prompt per 100k calls: $ 840.0
short prompt per 100k calls: $ 480.0
saving: $ 360.0
```

**Quiz:** Why does a long fixed prompt prefix matter for cost?

- [x] It is paid for on every single call
- [ ] It is free
- [ ] It is paid once per year
- [ ] It reduces output tokens

*Answer:* It is paid for on every single call. Fixed overhead multiplies with call volume.