Instrument 06 / cache economics

Know whether prompt caching saves money—or adds another bill.

Turn request volume, reusable tokens, current cache prices, and hit rate into monthly savings and one inspectable break-even threshold.

Model one month

Find the hit rate that pays for cache writes.

Nothing is submitted or stored. Every calculation runs in this browser, using the workload and current prices you provide.

Requests per monthExpected calls that contain the reusable prompt prefix. calls Reusable input tokens per requestOnly the stable prefix expected to qualify for caching. tok Uncached input price per 1MThe normal input rate used as the comparison baseline. $ Cache-read price per 1MThe price applied when the reusable prefix produces a hit. $ Cache-write price per 1MThe price applied when the reusable prefix misses and is written. $ Cache hit rateUse an observed production rate when one is available. %

Cache boundaries

Price misses.
A cache write can cost more than ordinary uncached input.

Measure hits.
The observed hit rate decides whether cheaper reads recover that premium.

Count stable input.
Changing content and output tokens do not belong in this model.

Transparent method

What this prompt caching calculator computes.

Uncached baseline multiplies requests by reusable input tokens and the normal uncached price. It asks what the same stable prefix would cost without caching.

Cache-read spend prices the hit share at the cache-read rate. Cache-write spend prices the miss share at the cache-write rate.

Monthly savings subtracts read and write spend from the uncached baseline. A negative result means the cache-write premium is larger than the cache-read savings at the stated hit rate.

Break-even hit rate solves for the hit share where cached and uncached input cost the same. Output tokens are excluded because prompt caching changes reusable input economics, not generated-token pricing.

This is a planning estimate, not a provider quote. Verify current prices, cache eligibility, minimum token requirements, cache lifetime, write behavior, invalidation rules, and any storage charges in the provider's current documentation.

Prompt cache guide

Four questions before treating cached tokens as savings.

A cheap cache-read price is only half the model. The write rate, prefix stability, eligibility rules, and real hit rate decide what reaches the invoice.

Read the full AI evaluation guide → Need the full input, output, and retry budget? →

How do you calculate prompt caching savings?

Price cache-hit reusable input tokens at the cache-read rate and cache-miss reusable input tokens at the cache-write rate. Subtract that total from the cost of pricing every reusable input token at the normal uncached rate.

What is a good prompt cache hit rate?

A useful hit rate is one above this workload's break-even threshold. The threshold depends on all three input prices, so there is no universal percentage that makes caching economical.

When can prompt caching cost more?

Caching can cost more when writes carry a premium and the hit rate is too low to recover it through cheaper reads. Short cache lifetimes and frequently changing prefixes can turn expected hits into writes.

Which input tokens belong in a cache calculation?

Include only the stable, reusable input prefix expected to qualify under current provider rules. Exclude changing user content, ineligible tokens, and output tokens.