✓ Last verified: 2026-07-28· Gemini/OpenAI explicit caching· report an issue →
Explicit context caching charges a per-hour storage fee for as long as the cached content is alive — on top of the cheaper reads — a tax the usual caching calculators ignore. This prices the storage fee, nets it against the read savings, and tells you the minimum reads and the maximum idle lifetime before storage eats the whole benefit.
—
net savings vs no cache
—storage fee
—break-even reads
—max idle lifetime
Net savings as you keep the cache alive longer
The read savings are fixed by how many reads you serve; only the storage fee grows with lifetime. Each row keeps the same reads but a longer time-to-live — watch the net flip from green to red once storage overtakes the savings. Your chosen lifetime is marked; the last profitable row is the sweet spot.
Lifetime
Storage fee
Read savings
Net
The tax the caching calculators leave out
Every prompt-caching guide sells you the same headline: cached reads cost a tenth of normal input, so caching cuts your bill by 90%. That is true for the reads. It is not the whole bill. Explicit context caching — the kind where you deliberately upload a big context and get back a handle to reuse — comes with a storage fee that runs on a clock, not on usage. Gemini charges roughly 4.50 dollars per million tokens per hour to keep 2.5 Pro content cached, so a 100,000-token context left alive for a day costs about 10.80 dollars in storage before anyone reads it. On a well-used cache that is fine, because the read savings dwarf it. On a cache that sits mostly idle — a document someone might ask about, a context wired to a generous time-to-live "just in case" — the storage line quietly becomes the biggest number on the invoice, and the cache loses money even though every read looks like a bargain. The two questions that actually decide it are how many reads you need before the savings cover the storage, and how long you can leave the cache alive before the meter overtakes those savings. This calculator answers both. If your context repeats on every call rather than living in an explicit cache, price the implicit case with the prompt caching savings calculator; to weigh a short versus long time-to-live against your hit rate, use the cache TTL & hit-rate calculator; and to see whether the write premium ever pays back, run the cache write break-even calculator.
It splits the cost into three lines. Storage is cached tokens (in millions) times lifetime in hours times the storage rate — it runs on the clock whether or not you read. The no-cache path is what you would pay without caching: every read pays the full input price for the whole context. The cache path is one write at the full input price, plus each read at the cheaper cached price, plus the storage fee. Net savings is the no-cache path minus the cache path. The break-even reads is the smallest number of reads at which the read savings cover the storage fee; the maximum idle lifetime is the longest you can keep the cache alive at your read count before storage overtakes the savings. It assumes one cache write over the window and ignores request output tokens, which are the same on both paths; real explicit-cache pricing and minimum-token floors vary by provider, so confirm on the official pricing page.
Frequently asked questions
What is context cache storage cost and why does it exist?
Explicit caching lets you upload a big, stable context once and reuse it at a cheap read price instead of re-sending it every call. But you also pay a per-hour storage fee for as long as the cache lives, used or not. Gemini bills around $4.50 per million tokens per hour on 2.5 Pro, so a 100k-token cache alive a day costs ~$10.80 in storage before any read. That storage line is the tax ordinary caching calculators leave out.
How is context cache storage priced?
Rate in dollars per million tokens per hour, times tokens cached (in millions), times hours alive. Gemini 2.5 Pro is ~$4.50/M/hour, Flash ~$1.00. A 100k-token cache (0.1M) for 24 hours on Pro is 0.1 × 24 × 4.50 = $10.80. It is driven by time and size, not usage — an idle cache still runs the meter, the opposite of reads, which cost only when used.
How many reads before caching a big context pays off?
Enough that the per-read savings (full input price minus cached price) exceed the fixed storage fee. For a 100k-token Gemini Pro context alive a day, that is roughly 117 reads before the savings cover the $10.80 storage — below that, caching costs more than paying full price each call. The calculator solves this minimum for your own size, prices and lifetime.
When does an idle context cache start losing money?
Once its storage fee outgrows the read savings already banked. Storage accrues every hour, so for a given read count there is a maximum lifetime past which the cache flips from saving to burning. At 200 reads on a 100k Gemini Pro cache, the savings absorb ~41 hours of storage; keep it alive longer and you pay to store content whose benefit you already collected. Match the cache's time-to-live to real traffic, not caution.