Free tool / cost

What would caching save you?

Enter how your prompts look. See the monthly saving from prompt caching on every model that supports it.

No sign-up / runs in your browser

Cache the fixed part, pay less.

What does one call look like?

System prompt, tool definitions, fixed documents.

User message and new tool results.

%

Caches expire after a few minutes without use.

Biggest saving

$9,360 a month

Claude Fable 5.1, 53% less

Typical saving

53%

Median across models with a cache price

Cost with caching

$8,240

Claude Fable 5.1, down from $17,600

Models compared

35

Only models with a published cache price

ModelWithout cacheWith cacheYou save%
Claude Fable 5.1 Anthropic$17,600$8,240$9,36053%
GPT-5.5 OpenAI$9,200$4,880$4,32047%
Claude Opus 4.7 Anthropic$8,800$4,480$4,32049%
Claude Opus 5.5 Anthropic$7,040$3,392$3,64852%
GPT-5.6 OpenAI$7,040$3,584$3,45649%
Claude Sonnet 4.6 Anthropic$5,280$2,688$2,59249%
GPT-5.4 OpenAI$4,600$2,440$2,16047%
Claude Sonnet 5.5 Anthropic$3,520$1,696$1,82452%
Grok 4.5 xAI$3,200$1,568$1,63251%
GPT-4.1 OpenAI$3,360$1,920$1,44043%
o3 OpenAI$3,360$1,920$1,44043%
Gemini 3.5 Flash Google$2,760$1,464$1,29647%
Mistral Medium Mistral$2,640$1,344$1,29649%
DeepSeek V4 Pro DeepSeek$2,112$887$1,22558%
GPT-4o OpenAI$4,200$3,000$1,20029%
GPT-5 OpenAI$2,500$1,420$1,08043%
Gemini 2.5 Pro Google$2,500$1,420$1,08043%
Grok 4.3 xAI$1,900$892$1,00853%
Claude Haiku 4.5 Anthropic$1,760$896$86449%
o4-mini OpenAI$1,848$1,056$79243%
GPT-5.4 mini OpenAI$1,380$732$64847%
Gemini 3.8 Flash Google$1,320$672$64849%
Mistral Large Mistral$800$368$43254%
GPT-4.1 mini OpenAI$672$384$28843%
DeepSeek V4 Flash DeepSeek$504$222$28256%
Gemini 2.5 Flash Google$608$349$25943%
DeepSeek Chat DeepSeek$414$172$24258%
GPT-5 mini OpenAI$500$284$21643%
Gemini 3.1 Flash-Lite Google$460$244$21647%
GPT-5.4 nano OpenAI$372$199$17346%
Mistral Small Mistral$252$122$13051%
Claude Haiku 5.5 Anthropic$176$89.60$86.4049%
Gemini 2.5 Flash-Lite Google$168$81.60$86.4051%
GPT-4o mini OpenAI$252$180$72.0029%
GPT-5 nano OpenAI$100$56.80$43.2043%

List prices in US dollars, checked 17 October 2026. Some providers charge extra to write the cache or have minimum prompt sizes. This estimate leaves those out, so check your provider’s caching rules.

This is a planning estimate from public list prices. Cache write fees, minimum prompt sizes and expiry times differ by provider.

How it works

Three steps, no sign-up.

  1. 01

    Describe one call

    Enter the fixed part of your prompt, the part that changes, and the reply length.

  2. 02

    Set your cache hit rate

    Guess how many calls find the fixed start already cached. Busy apps hit more often.

  3. 03

    Compare models

    See the monthly cost with and without caching on every model that has a cache price.

Questions people ask

What is prompt caching?

Providers store the start of a prompt they have just processed. If your next call begins with the same text, that part is billed at a much lower cached price.

What do I need to do to hit the cache?

Keep the fixed text first and the changing text last. Dates, user names and ids near the top change the start and break the match.

Why is my real saving lower than the estimate?

Caches expire after a short idle time, some providers charge extra to write the cache, and short prompts may not qualify. Measure your real cache hits before you trust the number.

Is my data stored?

No. The numbers stay in your browser.

How do I see my real cache hits?

Uncoil shows tokens and cost for every call in a run, so you can see which calls paid full price.

More free tools

  • Agent log checker

    Paste a log or trace. See where it failed, what was slow and what wasted tokens.

  • Token counter

    Paste text. See the token count and price on GPT, Claude, Gemini and Llama.

  • Model price comparison

    Every model's price in one table, plus what your app costs per month on each.

  • Prompt cost checker

    Paste a prompt. Get a shorter, cheaper version and what it costs on each model.

See all 12 free tools

Now measure the real thing.

Send a first trace and see the true cost, speed and failures of every run.