Caching

Cache Hit Calculator

When is prompt caching worth the write premium? Compute the break-even for your workload.

Cached prefix tokens8,000

Fresh tokens per request2,000

Requests per hour20

Cache TTL

No cache / hr

$0.6000

With cache / hr

$0.5280

Savings

−12.0%

Monthly delta

−$51.84

Break-even

You need at least 1.39 cache reads per write for caching to pay off. At your settings, you get 1.7 reads per write → caching is worth it.

Multipliers reflect public 2026 pricing: Anthropic ephemeral cache write = 1.25× base / 1h cache write = 2×, read = 0.1×. OpenAI prompt caching is auto and effectively free to write, 0.1× to read. Real-world savings also depend on cache invalidation rate and whether prefix stays byte-identical between calls.