Prompt caching
Find how many reuses it takes for prompt caching to save money.
Per MTok: input $4, cache write $5, cache read $0.2. Catalog 2026-10-10.
System prompt, few-shot examples, or any reusable context.
Calls that read the cache after the first call writes it.
Caching economics
Break-even = (write - input) / (input - read). The first call pays the write premium; each reuse saves input minus read.
- Cache write
- 125% of input
- Cache read
- 5% of input
- Saved per reuse
- 95% of input
- Break-even
- 0.26 reuses (pays from reuse 1)
Cache it
Cache it. You save $0.90 (75%)
You save
$0.9075% over 6 calls
- Without caching
- $1.20
- With caching
- $0.30
Breakdown
- Cache write (first call)
- $0.25
- Cache read x 5
- $0.05
- Total with caching
- $0.30
- vs. input x 6
- $1.20
Cost over reuses
Lines cross at break-even, 0.26 reuses.
- Without caching
- With caching