Estimate the marginal and fully allocated cost of one prompt-cached task. Enter separate token prices for uncached input, cached input, and output, then allocate optional cache-storage expense across the tasks completed while the cache is retained. The breakdown shows where spend comes from and compares the cached scenario with processing the entire prompt at the uncached-input rate.
The main result is fully allocated cost per task. Savings compares this with billing all input tokens at the uncached input rate while keeping output and allocated storage unchanged.
Prices vary by provider, model, cache-write rules, retention duration, region, and date. Enter current contract prices; taxes, network fees, retries, and non-model infrastructure are excluded unless added separately.
For 800 uncached input tokens at $3/M, 3,200 cached tokens at $0.30/M, 400 output tokens at $15/M, and no storage charge:
Only if you incorporate them into the entered prices or storage cost. Provider billing models differ.
That allocates one hour of cache retention expense evenly across the tasks completed during that hour.
It prices all input tokens, including the cached prefix, at the entered uncached-input rate while leaving output and storage unchanged.
They are usually priced and generated separately, so this calculator applies the output rate to them regardless of cache use.
No. It projects entered per-task model and allocated cache costs only. Add hosting, observability, and other fixed costs separately.
| Variable | Meaning | Unit |
|---|---|---|
| U, C, O | Uncached, cached, and output tokens | tokens/task |
| Pᵤ, P꜀, Pₒ | Token prices by class | $/million tokens |
| S | Cache storage or retention cost | $/hour |
| H | Tasks sharing storage | tasks/hour |
| Cost | Fully allocated unit cost | $/task |