Prompt cache simulator

Change the settings and watch which requests reuse the cached prefix, and what it does to cost.

Shared prefix

system prompt + tools + reference docs

New text per request

the user's new message

Requests

Gap between requests

Cache lifetime Put a changing timestamp at the top of the prompt (the most common way to lose the cache)

Prefix written to cachePrefix read from cacheNew text (full price)

Illustrative arithmetic, not a benchmark. Cost is counted in "input-token equivalents" using the multipliers Anthropic publishes (write 1.25x for 5 minutes or 2x for 1 hour, read 0.1x); other providers and models differ, and output tokens and latency are not modelled. Provider documentation.