Costiva
Home New to AI
Tools

All 47 tools → Features Use cases Live demo Impact analytics Pricing Blog FAQ About
Sign in Get started free
Free tool

Prompt Caching Savings Analyzer

Paste a prompt and see exactly what restructuring it for caching would save — and the call volume below which caching costs more than it saves.

Get started
  • Detects the static/dynamic split itself — no need to already know which part of your prompt is cacheable.
  • Flags any variable-looking content hiding inside what would otherwise be a clean cacheable prefix, since a single stray token there breaks caching for the whole thing.
  • Shows the break-even call frequency below which caching this prompt costs more than it saves — most tools never mention this exists.
Get started

Frequently asked questions

None — this tool never calls a model to generate anything. It only counts real tokens for the model you select and prices the result against Costiva's admin-editable price list.

Every line of your pasted prompt is checked against a set of patterns for dynamic-looking content — template placeholders, angle-bracket variables, dates, and similar. The first line that matches becomes the boundary. To set the boundary yourself instead, paste <<<CACHE_BOUNDARY>>> on its own line where the variable part starts.

Writing a prompt to the cache costs more per token than a normal fresh request — Anthropic prices a cache write at a premium over the base input rate. That premium only pays for itself if the cache gets reused (read) enough times before it expires. Call rarely enough relative to the cache TTL and every "cached" call is actually a fresh write that never gets reused — the break-even frequency below is the exact point where that stops being true.

Anthropic caches a prompt prefix for 5 minutes by default, or up to 1 hour with extended caching (at a higher write price). Each cache read refreshes that window, so a prompt called more often than once per TTL effectively never falls out of cache — call it less often than that and every call pays the cache-write premium instead of the cache-read discount.