Prompt Caching Savings Calculator
Estimate how much prompt caching saves when a long shared prefix (system prompt, context, docs) is reused across many requests, at the provider’s cache discount.
How it works
Prefix cost = cached tokens × requests ÷ 1M × price · Savings = prefix cost × cache discount%
When many requests share a long, unchanging prefix — a big system prompt, few-shot examples or retrieved context — prompt caching lets the provider skip re-processing it and bills those tokens at a steep discount, often 75–90% off. The savings scale with both prefix length and request volume, so caching pays off most for high-traffic apps with heavy shared context and barely matters for one-off calls. Note this models only the cached prefix; the variable part of each prompt and the output tokens are billed at full price on top.
Worked example
A 5,000-token prefix reused across 100,000 requests is 500M prefix tokens/month. At $3 per million that is $1,500 without caching; a 90% cache discount saves $1,350, leaving $150 — about $16,200 saved per year.
Frequently asked questions
What can actually be cached?
The stable prefix of your prompt — system instructions, tool definitions, few-shot examples and retrieved documents that do not change between calls. The dynamic tail (the user’s latest message) is not cached and is billed normally.
Is there a cost to writing the cache?
Some providers charge a small premium to create or refresh a cache entry, then discount reads heavily. If your prefix is reused many times before it expires, the read savings dwarf the write cost — this calculator models the read savings, so treat it as a slight upper bound.
Why doesn’t caching help my app?
If each request has a different prefix, or you send few requests before the cache expires (often a few minutes), there is little to reuse. Caching rewards high volume over a shared, long-lived prefix.
Related calculators
Educational estimate — real costs and memory depend on your provider, framework, model architecture and pricing, which change often. Verify against your own bills and profiling before committing budget.