Recommended Free Tools
Anthropic prompt caching can lower API costs when requests repeatedly include the same substantial prefix—such as system instructions, tool definitions, examples, or a long document. Mark the stable content for caching, keep changing content after it, and confirm reuse in the response’s cache-usage fields. A cache marker alone does not guarantee a hit, and cache writes cost more than ordinary input.
How Anthropic prompt caching works
Prompt caching lets Claude reuse a matching prefix of a request up to a cache breakpoint instead of processing that material as ordinary input on every call. Suitable repeated content can include system instructions, tool definitions, text, documents or images in user turns, and prior tool-use or tool-result content. It is most useful when a large, stable context recurs.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
The Claude AI Advanced Handbook: Model and Effort Economics for Claude Opus 5: Real Cost Per Task,... | $9.99 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
Anthropic describes the feature as reducing costs and latency by reusing previously processed portions of a prompt across API calls. See the official prompt caching guide.
Choose automatic caching or explicit breakpoints
Automatic caching
For a simple starting point, add cache_control: {"type": "ephemeral"} at the request’s top level. Anthropic says this automatically places the breakpoint on the last cacheable block and moves it as conversation history grows.
Explicit breakpoints
Place cache_control on selected content blocks when you need to choose exactly which prefix sections to cache. This is useful when stable instructions and changing retrieval context should have separate boundaries. Anthropic supports up to four breakpoints; adding them does not itself increase charges.
Put a breakpoint after the last block that remains identical between requests and before variable material such as the current user message. If cached content or relevant request settings change, some or all of the prefix may no longer match.
Set up a request for reuse
- Find the repeated material. Identify large content that recurs unchanged, such as system instructions, tool definitions, examples, or a long document.
- Mark the stable prefix. Use automatic caching for a straightforward request or conversation; use explicit breakpoints when parts of the prompt change at different rates.
- Keep changing content after the breakpoint. Put per-request details, timestamps, and the incoming user message after the stable prefix. Avoid changing the cached content or tool definitions if you expect a match.
- Choose a cache lifetime. The default five-minute TTL suits requests that recur within five minutes. Consider the one-hour TTL only when the longer window is useful enough to justify its higher write premium.
- Check for reuse in the response. Inspect
cache_creation_input_tokensandcache_read_input_tokens, not onlyinput_tokens.
Anthropic says a cache entry becomes available after the first response begins. Concurrent requests sent before that point may not receive a cache hit.
Understand TTLs and cache pricing
Anthropic’s pricing documentation, checked on October 7, 2026, gives these standard multipliers. They are not a substitute for checking the current rate for your specific model: prices are model-specific, can change, and Anthropic documents model-specific exceptions to the standard cache-read multiplier. See the official pricing page.
| Decision factor | Five-minute TTL | One-hour TTL |
|---|---|---|
| Standard cache-write price | 1.25× base input price, per Anthropic’s pricing documentation checked October 7, 2026 | 2× base input price, per Anthropic’s pricing documentation checked October 7, 2026 |
| Standard cache-read price | 0.1× base input price, subject to model-specific exceptions | 0.1× base input price, subject to model-specific exceptions |
| When it may fit | Repeated use within five minutes; reuse refreshes the cache without additional cost | Reuse gaps longer than five minutes but under an hour, or operational needs that justify the higher write price |
| Main trade-off | A long response consumes part of the five-minute window | The cache write has a higher premium |
Anthropic measures the five-minute TTL from the start of the request that writes or reads the entry, not from response completion. Time spent generating a long answer therefore reduces the remaining time for reuse. Anthropic says reuse refreshes the cache without additional cost; a five-minute cache can be refreshed this way when prompts recur more frequently than every five minutes.
At the standard 0.1× cache-read multiplier, Anthropic says a five-minute cache write can break even after one cache read, while a one-hour write can break even after two reads. These are comparisons of write premiums with reads at 10% of base input cost, not guaranteed savings. Prompt size, model rate, hit rate, TTL, and expiration timing all affect the result.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Verify cache use and troubleshoot misses
Read the right usage fields
Anthropic defines total input tokens as input_tokens + cache_creation_input_tokens + cache_read_input_tokens. The input_tokens field alone represents only the uncached portion after the last breakpoint, so it is not a complete measure of prompt input.
If neither cache count appears
- Check that the prompt meets the model’s minimum cacheable length; the minimum varies by model.
- Confirm that the content before the breakpoint and relevant request settings match the earlier request.
- Check whether the next request arrived before the first response began, when the cache entry may not yet be available.
- Consult the current documentation for model- and platform-specific usage fields and requirements.
Check platform and model support
Anthropic’s guide lists active Claude models and the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry as supported. Minimum cacheable lengths, usage field names, and setup instructions may differ by model or hosting platform, so follow the provider-specific guidance for your deployment.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
