Prompt-caching & batch savings calculator
Two of the biggest levers on an LLM bill are already built into the APIs: prompt caching (cheap reads of the tokens that repeat every request) and the batch API (a discount for work you don't need back in real time). 10 tracked models publish a caching discount and 11 publish a batch discount — set your volume below to see what each is worth on your workload.
$0.05 in · $0.40 out per 1M tokens
The share of your input tokens that repeat across requests (system prompts, long context, few-shot examples). Only the cached share gets the caching discount.
Discount rates are each provider's own published prompt-caching and batch-API pricing, applied to your stated volume. Caching and batch are shown separately because a provider's batch queue usually doesn't serve real-time cache reads — combining both would overstate the saving. Estimates, not a bill; confirm current rates with the provider.
When prompt caching pays off
Caching rewards repetition. A long system prompt, a fixed instruction block or a large context reused across many calls can land a high cache-hit rate, so most of your input tokens bill at the discounted read price. Short, one-off prompts see little from it.
When the batch API pays off
Batch trades latency for price: submit the work and collect results within the provider's window (often up to 24h). Ideal for evals, backfills, bulk classification and overnight jobs — anything a user isn't waiting on.