Prompt caching bills repeated input tokens at the cache-read rate instead of the standard input rate. The table below shows the effective input price and monthly cost at different cache hit rates for a 10M-input / 1M-output token month.
10,000,000 input tokens and 1,000,000 output tokens per month. Output pricing is unchanged.
| Cache hit rate | Effective input / 1M | Monthly input cost | Monthly total | Saving vs no cache |
|---|---|---|---|---|
| 0% | $0.100 | $1.00 | $1.50 | $0.00 |
| 50% | $0.055 | $0.55 | $1.05 | $0.45 |
| 75% | $0.033 | $0.33 | $0.82 | $0.68 |
| 90% | $0.019 | $0.19 | $0.69 | $0.81 |
| 95% | $0.015 | $0.15 | $0.65 | $0.85 |
Providers only bill the cache-read rate when repeated content qualifies for caching and the cache is still warm. This page is an arithmetic estimate from stored catalog rates, not a guarantee of provider billing behaviour. Confirm cache rules and minimum token thresholds in the provider documentation.