Prompt Caching Savings Calculator
See when a long repeated system prompt is worth caching.
Estimate your workload
Change the assumptions on the left; read the modeled result on the right.
How to read this estimate
What this calculator is for
Prompt caching matters most when a large, stable prefix is reused across many requests. The calculator separates cacheable and changing input, then applies an estimated hit rate.
How to read the result
Input cost with cache is the modeled monthly input-token charge. Without caching is the baseline. Estimated savings is their difference, and savings rate expresses that difference as a share of the baseline. This calculator excludes output-token charges.
The math
Input cost with caching = uncached tokens at the standard rate + cacheable tokens split across misses at standard rate and hits at the cached-input rate.
Pricing sources
Rates use the published TokenCOGS pricing registry and its September 16, 2026 fallback snapshot. Live values may differ; see methodology and source limitations before procurement.
Planning limitations
Actual invoices can differ because providers may apply long-context pricing, caching rules, batch rates, tool-call fees, regional processing, negotiated contracts, taxes, or future pricing changes. Treat the output as a planning estimate.
Use it to create scenarios and catch order-of-magnitude mistakes. Before committing spend, verify the selected model's live provider pricing and compare the estimate with actual usage telemetry.
What happens if the model has no published cached-input rate?
The tool shows an unavailable-rate notice instead of inventing a cached rate. Choose another model or check the provider terms.
What content is a good caching candidate?
Large repeated system prompts, tool definitions, policies, schemas and stable document context are common candidates when a provider supports prompt caching.