Now the money. An agent resends its whole conversation every turn, which would be ruinous if every token cost full price every time.
Prompt caching
The provider keeps the processed start of your request for a few minutes. If your next request begins with exactly the same bytes, that part costs a fraction of the price. One changed byte invalidates everything after it.
For long agent sessions this usually matters more than any wording trick. A timestamp at the top of a system prompt breaks every hit, and because saving to the cache costs extra, you then pay more than with no caching at all.
“Move the date out of the system prompt and into the last user message. Then send two identical requests and show me usage.cache_read_input_tokens from the second: it must be above zero.”