Question 1

What is context cache storage cost and why does it exist?

Accepted Answer

Explicit context caching lets you upload a big, stable chunk of context once — a long document, a system prompt, a codebase — and reuse it across many requests without re-sending and re-paying for those tokens each time. The reads are cheap, often a quarter or a tenth of the normal input price. But there is a second charge most people miss: a per-hour storage fee that you pay for as long as the cached content stays alive, whether or not you read it. Gemini, for example, bills roughly 4.50 dollars per million tokens per hour to keep 2.5 Pro content cached, so a 100,000-token cache kept alive for a day costs about 10.80 dollars in storage before a single read. That storage line is the tax the ordinary prompt-caching calculators leave out, and on a large context with sparse reads it can be the biggest number on the bill. This calculator prices it directly and nets it against the read savings.

Question 2

How is context cache storage priced?

Accepted Answer

It is a rate in dollars per million tokens per hour, multiplied by how many tokens you cache and how many hours you keep them alive. So storage cost equals cached tokens, in millions, times lifetime in hours, times the rate. On Gemini 2.5 Pro that rate is around 4.50 dollars per million tokens per hour; on the cheaper Flash tier it is closer to 1.00 dollar. A 100,000-token cache is 0.1 million tokens, so keeping it alive 24 hours on Pro costs 0.1 times 24 times 4.50, which is 10.80 dollars. The key thing to notice is that this cost is driven by time and size, not by usage — an idle cache that no request touches still runs the meter. That is the opposite of the reads, which cost you only when you use them, and it is why a big cache that sits mostly idle can lose money even though every individual read looks cheap. This calculator separates the two so you can see which one dominates your bill.

Question 3

How many reads do I need before caching a big context pays off?

Accepted Answer

That is the break-even question, and storage is what makes it non-trivial. Without caching, every request pays the full input price for the whole context. With explicit caching you pay a cheap read each time, plus the fixed storage fee for the lifetime. The reads save you the gap between full and cached input price on every call, but those savings only start to matter once they exceed the storage fee you are paying regardless. For a 100,000-token context on Gemini Pro kept alive a day, you need roughly 117 reads before the read savings cover the 10.80-dollar storage bill — below that, caching costs more than just paying full price each time. This calculator solves that minimum-reads number for your own context size, prices and lifetime, so you know whether your traffic is dense enough to justify the cache or whether you are paying storage rent on a context nobody is reading often enough.

Question 4

When does an idle context cache start losing money?

Accepted Answer

The moment its storage fee grows past the read savings it has already banked. Because storage accrues with every hour the cache is alive, there is a maximum lifetime for any given number of reads beyond which the cache flips from saving money to burning it. If you serve 200 reads against a 100,000-token Gemini Pro cache, the read savings can absorb roughly 41 hours of storage before the meter overtakes them; keep the cache alive past that and you are paying to store content whose savings you have already collected. This is the classic trap of explicit caching: a team caches a large context, wires it to a long time-to-live to be safe, and then serves it far less than expected, so the storage tax quietly outgrows the benefit. This calculator prints that maximum idle lifetime alongside the break-even reads, so you can match your cache's time-to-live to your real traffic instead of leaving it running out of caution.

Calculadora de custo de armazenamento de cache de contexto

Economia líquida à medida que você mantém o cache ativo por mais tempo

O imposto que as calculadoras de cache deixam de fora

Como funciona esta calculadora

Perguntas frequentes

Qual é o custo de armazenamento em cache de contexto e por que ele existe?

Qual é o preço do armazenamento em cache de contexto?

Quantas leituras antes de armazenar em cache um grande contexto compensam?

Quando um cache de contexto ocioso começa a perder dinheiro?