Prompt Caching Calculator
See how much prompt caching actually saves. Enter your prefix size, request rate and cache TTL to get a monthly figure and the break-even verdict for your provider.
5-min cache write is 1.25x base, reads are 0.1x (90% off).
System prompt, tools, fixed docs.
The user message, never cached.
How often this prefix is reused.
How long a warm cache lives.
Reads per write are derived from your request rate and TTL, assuming evenly spaced calls. Fresh tokens always bill at full price. Prices are representative 2026 figures, set “Custom” for your exact contract.
What moves the saving
Three things decide whether caching pays off. The calculator captures all three; here is what each one does.
Read discount
Cached input tokens bill at about 10% of the standard price. The bigger and more repeated your prefix, the more each hit saves.
Write premium
Some providers charge 1.25x (5-min) or 2x (1-hour) to write a cache. You need enough reads per write to clear that premium.
Prefix stability
A cache hit needs a byte-identical prefix. One timestamp in the wrong place drops your hit rate to zero and you pay to re-cache every call.
Caching is one of three levers for cutting your model bill. Read the full guide to prompt caching and the broader playbook on LLM cost optimization.
Prompt caching calculator FAQ
How much does prompt caching save?+
Cache reads are billed at roughly 10% of the standard input price across the major providers, so the repeated span of a prompt costs about 90% less once it is cached. The exact saving depends on your cache hit rate: the calculator multiplies your cached-prefix cost by the read discount on hits and the write premium on misses to give a monthly figure for your own numbers.
What is the break-even for prompt caching?+
Providers that charge a write premium (Anthropic and GPT-5.6 add 1.25x on a cache write) only save money once enough requests read the cache to outweigh that premium. With a 1.25x write and a 0.1x read, break-even lands around a 22% hit rate. Above that you save, below it caching costs more. Providers with no write fee (DeepSeek, older OpenAI, Gemini implicit) save from the first read.
What counts as the cacheable prefix?+
The stable content at the start of every request: your system prompt, tool and function definitions, few-shot examples, and any fixed reference documents. It has to be identical byte-for-byte across requests to hit the cache. The variable part, the user message, always sits at the end and is never cached, which is why it does not affect the saving.
Is a higher cache hit rate always better?+
Yes, for cost. Every hit is billed at the read discount instead of full price, and on a 5-minute cache each hit also refreshes the timer for free. The usual reason hit rates are low is a dynamic element (a timestamp, session ID or rotating instruction) buried in the prefix, which changes the hash and forces a fresh write every request. Move anything variable to the end of the prompt.
Which providers does this calculator cover?+
It ships with presets for Anthropic (Claude Opus, Sonnet, Haiku, 5-minute and 1-hour caches), OpenAI (GPT-5.6+ explicit and legacy automatic caching), Google Gemini (implicit and explicit) and DeepSeek, using their published 2026 multipliers. Pick "Custom" to enter the exact base price and write/read multipliers from your own contract.
Cut your AI bill without cutting quality
The calculator shows the prize. We build the caching, routing and cost instrumentation that captures it. Book a free cost review.