Free cloud & LLM pricing tools — AWS EC2 + LLM API Pricing Explorer

Cache Write

The one-time fee some providers charge to store a prompt prefix in the cache before you can read it back cheaply.

A cache write is the cost of putting a prompt prefix into the cache so later requests can reuse it at the cheap cached-input rate. Anthropic, for example, charges a cache-write price above the standard input rate — its 5-minute cache write is 1.25× the input rate, and a 1-hour write is 2× — in exchange for cache reads at roughly one-tenth of input.

The economics are a break-even calculation. You pay the write premium once, then save on every subsequent read within the cache’s lifetime. If the same prefix is reused many times before it expires, the write fee is trivial against the read savings; if it is reused only once or twice, the write can cost more than it saves. High reuse is what makes caching pay off.

This explorer surfaces the 5-minute cache-write rate for providers that publish it (chiefly Anthropic) as its own figure on the model page. Providers that fold caching into a single cached-token rate, or that do not publish a write fee, show a dash — never a guessed number. Always confirm cache lifetimes and write fees against the provider’s own docs before sizing a caching strategy.

FAQ

Why is cache write more expensive than normal input?

Writing to the cache stores extra state, so providers charge a premium — Anthropic’s 5-minute write is 1.25× the input rate. You recoup it through cheap reads on later calls.

Do all providers charge a cache-write fee?

No. Some publish only a single cached-input (read) rate, and others do not document caching pricing at all. This explorer shows a write fee only where the provider lists one.

More LLM pricing terms

← Back to the LLM pricing hub