Prompt caching

Prompt caching optimizes your API usage by allowing resuming from specific prefixes in your prompts. This significantly reduces processing time and costs for repetitive tasks or prompts with consistent elements.

There are two ways to enable prompt caching:

• Automatic caching: Add a single cache_control field at the top level of your request. The system automatically applies the cache breakpoint to the last cacheable block and moves it forward as conversations grow. Best for multi-turn conversations where the growing message history should be cached automatically.
• Explicit cache breakpoints: Place cache_control directly on individual content blocks for fine-grained control over exactly what gets cached.

The simplest way to start is with automatic caching:

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-5",
    max_tokens=1024,
    cache_control={"type": "ephemeral"},
    system="You are an AI assistant tasked with analyzing literary works. Your goal is to provide insightful commentary on themes, characters, and writing style.",
    messages=[
        {
            "role": "user",
            "content": "Analyze the major themes in 'Pride and Prejudice'.",
        }
    ],
)
print(response.usage.model_dump_json())

With automatic caching, the system caches all content up to and including the last cacheable block. On subsequent requests with the same prefix, cached content is reused automatically.

How prompt caching works

When you send a request with prompt caching enabled:

1. The system checks if a prompt prefix, up to a specified cache breakpoint, is already cached from a recent query.
2. If found, it uses the cached version, reducing processing time and costs.
3. Otherwise, it processes the full prompt and caches the prefix once the response begins.

This is especially useful for:

• Prompts with many examples
• Large amounts of context or background information
• Repetitive tasks with consistent instructions
• Long multi-turn conversations

By default, the cache has a 5-minute lifetime. The cache is refreshed for no additional cost each time the cached content is used.

The lifetime is measured from the start of the request that writes or reads the cache entry, not from the end of its response. Time spent generating a response counts against the lifetime: if a response takes 4 minutes to stream, a follow-up request that reuses the same cached prefix must start within about 1 minute of that response completing.

Pricing

Prompt caching introduces a new pricing structure. The following table shows the price per million tokens for each supported model:

^{1 Cache hits and refreshes on Claude Fable 5.1 and Claude Mythos 5.1 are priced at 0.025x the base input price. All other models use the standard 0.1x multiplier.}

Supported models

Prompt caching (both automatic and explicit) is supported on all active Claude models.

Automatic caching

Automatic caching is the simplest way to enable prompt caching. Instead of placing cache_control on individual content blocks, add a single cache_control field at the top level of your request body. The system automatically applies the cache breakpoint to the last cacheable block.

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-5",
    max_tokens=1024,
    cache_control={"type": "ephemeral"},
    system="You are a helpful assistant that remembers our conversation.",
    messages=[
        {"role": "user", "content": "My name is Alex. I work on machine learning."},
        {
            "role": "assistant",
            "content": "Nice to meet you, Alex! How can I help with your ML work today?",
        },
        {"role": "user", "content": "What did I say I work on?"},
    ],
)
print(response.usage.model_dump_json())

How automatic caching works in multi-turn conversations

With automatic caching, the cache point moves forward automatically as conversations grow. Each new request caches everything up to the last cacheable block, and previous content is read from cache.

The cache breakpoint automatically moves to the last cacheable block in each request, so you don't need to update any cache_control markers as the conversation grows.

TTL support

By default, automatic caching uses a 5-minute TTL. You can specify a 1-hour TTL at 2x the base input token price:

{ "cache_control": { "type": "ephemeral", "ttl": "1h" } }

Combining with block-level caching

Automatic caching is compatible with explicit cache breakpoints. When used together, the automatic cache breakpoint uses one of the 4 available breakpoint slots.

This lets you combine both approaches. For example, use an explicit breakpoint to cache your system prompt, while automatic caching handles the conversation:

{
  "model": "claude-opus-5",
  "max_tokens": 1024,
  "cache_control": { "type": "ephemeral" },
  "system": [
    {
      "type": "text",
      "text": "You are a helpful assistant.",
      "cache_control": { "type": "ephemeral" }
    }
  ],
  "messages": [{ "role": "user", "content": "What are the key terms?" }]
}

What stays the same

Automatic caching uses the same underlying caching infrastructure. Pricing, minimum token thresholds, context ordering requirements, and the 20-block lookback window all apply the same as with explicit breakpoints.

Edge cases

• If the last block already has an explicit cache_control with the same TTL, automatic caching is a no-op.
• If the last block has an explicit cache_control with a different TTL, the API returns a 400 error.
• If 4 explicit block-level breakpoints already exist, the API returns a 400 error (no slots left for automatic caching).
• If the last block is not eligible as an automatic cache breakpoint target, the system silently walks backwards to find the nearest eligible block. If none is found, caching is skipped.

Explicit cache breakpoints

For more control over caching, you can place cache_control directly on individual content blocks. This is useful when you need to cache different sections that change at different frequencies, or need fine-grained control over exactly what gets cached.

Structuring your prompt

Place static content (tool definitions, system instructions, context, examples) at the beginning of your prompt. Mark the end of the reusable content for caching using the cache_control parameter.

Cache prefixes are created in the following order: tools, system, then messages. This order forms a hierarchy where each level builds upon the previous ones.

How automatic prefix checking works

You can use just one cache breakpoint at the end of your static content, and the system will automatically find the longest prefix that a prior request already wrote to the cache. Understanding how this works helps you optimize your caching strategy.

Three core principles:

1. Cache writes happen only at your breakpoint. Marking a block with cache_control writes exactly one cache entry: a hash of the prefix ending at that block. The system does not write entries for any earlier position. Because the hash is cumulative, covering everything up to and including the breakpoint, changing any block at or before the breakpoint produces a different hash on the next request.

2. Cache reads look backward for entries that prior requests wrote. On each request the system computes the prefix hash at your breakpoint and checks for a matching cache entry. If none exists, it walks backward one block at a time, checking whether the prefix hash at each earlier position matches something already in the cache. It is looking for prior writes, not for stable content.

3. The lookback window is 20 blocks. The system checks at most 20 positions per breakpoint, counting the breakpoint itself as the first. If the system finds no matching entry in that window, checking stops (or resumes from the next explicit breakpoint, if any). On the Claude API, a run of consecutive tool_use blocks counts as one position, and so does a run of consecutive tool_result blocks, so a turn with many parallel tool calls doesn't push the previous request's entry out of the window on its own.

Example: Lookback in a growing conversation

You append new blocks each turn and set cache_control on the final block of each request:

• Turn 1: 10 blocks, breakpoint on block 10. No prior cache entries exist. The system writes an entry at block 10.
• Turn 2: 15 blocks, breakpoint on block 15. Block 15 has no entry, so the system walks back to block 10 and finds the turn-1 entry. Cache hit at block 10; the system processes only blocks 11 through 15 fresh and writes a new entry at block 15.
• Turn 3: 35 blocks, breakpoint on block 35. The system checks 20 positions (blocks 35 through 16) and finds nothing. The turn-2 entry at block 15 is one position outside the window, so there is no cache hit. Adding a second breakpoint at block 15 starts a second lookback window there, which finds the turn-2 entry.

Common mistake: Breakpoint on content that changes every request

Your prompt has a large static system context (blocks 1 through 5) followed by a per-request block containing a timestamp and the user message (block 6). You set cache_control on block 6:

• Request 1: Cache write at block 6. The hash includes the timestamp.
• Request 2: The timestamp differs, so the prefix hash at block 6 differs. The lookback walks through blocks 5, 4, 3, 2, and 1, but the system never wrote an entry at any of those positions. No cache hit. You pay for a fresh cache write on every request and never get a read.

The lookback does not find stable content behind your breakpoint and cache it. It finds entries that prior requests already wrote, and writes happen only at breakpoints. Move cache_control to block 5, the last block that stays the same across requests, and every subsequent request reads the cached prefix. Automatic caching hits the same trap: it places the breakpoint on the last cacheable block, which in this structure is the one that changes every request, so use an explicit breakpoint on block 5 instead.

Key takeaway: Place cache_control on the last block whose prefix is identical across the requests you want to share a cache. In a growing conversation the final block works as long as each turn adds fewer than 20 blocks: earlier content never changes, so the next request's lookback finds the prior write. For a prompt with a varying suffix (timestamps, per-request context, the incoming message), place the breakpoint at the end of the static prefix, not on the varying block.

When to use multiple breakpoints

You can define up to 4 cache breakpoints if you want to:

• Cache different sections that change at different frequencies (for example, tools rarely change, but context updates daily)
• Have more control over exactly what gets cached
• Ensure a cache hit when a growing conversation pushes your breakpoint 20 or more blocks past the last cache write

Understanding cache breakpoint costs

Cache breakpoints themselves don't add any cost. You are only charged for:

• Cache writes: When new content is written to the cache (25% more than base input tokens for 5-minute TTL)
• Cache reads: When cached content is used (10% of base input token price, or 2.5% on Claude Fable 5.1 and Claude Mythos 5.1)
• Regular input tokens: For any uncached content

Adding more cache_control breakpoints doesn't increase your costs - you still pay the same amount based on what content is actually cached and read. The breakpoints give you control over what sections can be cached independently.

Caching strategies and considerations

Cache limitations

On the Claude API, Claude Platform on AWS, Google Cloud, and Microsoft Foundry, the minimum cacheable prompt length is:

• 512 tokens for Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5, Claude Fable 5, and Claude Mythos 5
• 2,048 tokens for Claude Mythos Preview and Claude Opus 4.7
• 4,096 tokens for Claude Opus 4.6 and Claude Opus 4.5
• 1,024 tokens for Claude Opus 4.8, Claude Sonnet 5, Claude Sonnet 4.6, Claude Sonnet 4.5, Claude Opus 4.1 (retired, except on Bedrock and Google Cloud), Claude Opus 4 (retired, except on Google Cloud), and Claude Sonnet 4 (retired, except on Bedrock and Google Cloud)
• 4,096 tokens for Claude Haiku 4.5
• 2,048 tokens for Claude Haiku 3.5 (retired, except on Bedrock and Google Cloud)

These minimums apply on every platform where each model is available.

Shorter prompts cannot be cached, even if marked with cache_control. Any requests to cache fewer than this number of tokens will be processed without caching, and no error is returned. To verify whether a prompt was cached, check the response usage fields: if both cache_creation_input_tokens and cache_read_input_tokens are 0, the prompt was not cached (likely because it did not meet the minimum length requirement).

If your prompt falls just short of the minimum for your model and platform, expanding the cached content to reach the threshold is often worthwhile. Cache reads cost significantly less than uncached input tokens, so reaching the minimum can reduce costs for frequently reused prompts.

For concurrent requests, note that a cache entry only becomes available after the first response begins. If you need cache hits for parallel requests, wait for the first response before sending subsequent requests.

Currently, "ephemeral" is the only supported cache type, which by default has a 5-minute lifetime.

What can be cached

Most blocks in the request can be cached. This includes:

• Tools: Tool definitions in the tools array
• System messages: Content blocks in the system array
• Text messages: Content blocks in the messages.content array, for both user and assistant turns
• Images & Documents: Content blocks in the messages.content array, in user turns
• Tool use and tool results: Content blocks in the messages.content array, in both user and assistant turns

Each of these elements can be cached, either automatically or by marking them with cache_control.

What cannot be cached

While most request blocks can be cached, there are some exceptions:

• Thinking blocks cannot be cached directly with cache_control. However, thinking blocks CAN be cached alongside other content when they appear in previous assistant turns. When cached this way, they DO count as input tokens when read from cache.

• Sub-content blocks (like citations) themselves cannot be cached directly. Instead, cache the top-level block.

  In the case of citations, the top-level document content blocks that serve as the source material for citations can be cached. This allows you to use prompt caching with citations effectively by caching the documents that citations will reference.

• Empty text blocks cannot be cached.

What invalidates the cache

Modifications to cached content can invalidate some or all of the cache.

As described in Structuring your prompt, the cache follows the hierarchy: tools → system → messages. Changes at each level invalidate that level and all subsequent levels.

The following table shows which parts of the cache are invalidated by different types of changes. ✘ indicates that the cache is invalidated, while ✓ indicates that the cache remains valid.

Tracking cache performance

Monitor cache performance using these API response fields, within usage in the response (or message_start event if streaming):

• cache_creation_input_tokens: Number of tokens written to the cache when creating a new entry.
• cache_read_input_tokens: Number of tokens retrieved from the cache for this request.
• input_tokens: Number of input tokens which were not read from or used to create a cache (that is, tokens after the last cache breakpoint).

Caching with thinking blocks

When using thinking with prompt caching, thinking blocks have special behavior:

Automatic caching alongside other content: While thinking blocks cannot be explicitly marked with cache_control, they get cached as part of the request content when you make subsequent API calls with tool results. This commonly happens during tool use when you pass thinking blocks back to continue the conversation.

Input token counting: When thinking blocks are read from cache, they count as input tokens in your usage metrics. This is important for cost calculation and token budgeting.

Cache invalidation patterns:

• Cache remains valid when only tool results are provided as user messages
• On Opus 4.5+ and Sonnet 4.6+, thinking blocks are preserved by default even when non-tool-result user content is added, so the cache remains valid
• On earlier Opus/Sonnet models and all Haiku models, cache gets invalidated when non-tool-result user content is added, causing all previous thinking blocks to be stripped from context
• This caching behavior occurs even without explicit cache_control markers

For more details on cache invalidation, see What invalidates the cache.

Example with tool use:

Request 1: User: "What's the weather in Paris?"
Response: [thinking_block_1] + [tool_use block 1]

Request 2:
User: ["What's the weather in Paris?"],
Assistant: [thinking_block_1] + [tool_use block 1],
User: [tool_result_1, cache=True]
Response: [thinking_block_2] + [text block 2]
# Request 2 caches its request content (not the response)
# The cache includes: user message, thinking_block_1, tool_use block 1, and tool_result_1

Request 3:
User: ["What's the weather in Paris?"],
Assistant: [thinking_block_1] + [tool_use block 1],
User: [tool_result_1, cache=True],
Assistant: [thinking_block_2] + [text block 2],
User: [Text response, cache=True]
# On earlier Opus/Sonnet and all Haiku models, non-tool-result user block causes prior thinking blocks to be stripped; on Opus 4.5+/Sonnet 4.6+ they are kept

On earlier Opus/Sonnet models and all Haiku models, all previous thinking blocks are removed from context at this point. On Opus 4.5+ and Sonnet 4.6+, prior thinking blocks are kept by default and remain part of the cached prefix.

For more detailed information, see Thinking and prompt caching.

Cache storage and sharing

• Organization and workspace isolation: Caches are isolated between organizations. Different organizations never share caches, even if they use identical prompts. Caches are also isolated per workspace within an organization on the Claude API, Claude Platform on AWS, and Microsoft Foundry; Bedrock and Google Cloud use organization-level isolation only.

• Exact matching: Cache hits require 100% identical prompt segments, including all text and images up to and including the block marked with cache control.

• Output token generation: Prompt caching has no effect on output token generation. The response you receive is identical to what you would get if prompt caching were not used.

Best practices for effective caching

To optimize prompt caching performance:

• Start with automatic caching for multi-turn conversations. It handles breakpoint management automatically.
• Use explicit block-level breakpoints when you need to cache different sections with different change frequencies.
• Cache stable, reusable content like system instructions, background information, large contexts, or frequent tool definitions.
• Place cached content at the prompt's beginning for best performance.
• Use cache breakpoints strategically to separate different cacheable prefix sections.
• Place the breakpoint on the last block that stays identical across requests. For a prompt with a static prefix and a varying suffix (timestamps, per-request context, the incoming message), that is the end of the prefix, not the varying block.
• Regularly analyze cache hit rates and adjust your strategy as needed.

Optimizing for different use cases

Tailor your prompt caching strategy to your scenario:

• Conversational agents: Reduce cost and latency for extended conversations, especially those with long instructions or uploaded documents.
• Coding assistants: Improve autocomplete and codebase Q&A by keeping relevant sections or a summarized version of the codebase in the prompt.
• Large document processing: Incorporate complete long-form material including images in your prompt without increasing response latency.
• Detailed instruction sets: Share extensive lists of instructions, procedures, and examples to fine-tune Claude's responses. Developers often include an example or two in the prompt, but with prompt caching you can get even better performance by including 20+ diverse examples of high quality answers.
• Agentic tool use: Enhance performance for scenarios involving multiple tool calls and iterative code changes, where each step typically requires a new API call.
• Talk to books, papers, documentation, podcast transcripts, and other longform content: Bring any knowledge base alive by embedding the entire document(s) into the prompt, and letting users ask it questions.

Troubleshooting common issues

If experiencing unexpected behavior:

• Ensure cached sections are identical across calls. For explicit breakpoints, verify that cache_control markers are in the same locations
• Check that calls are made within the cache lifetime (5 minutes by default)
• Verify that tool_choice, image usage, the thinking configuration, and output_config.effort remain consistent between calls
• Validate that you are caching at least the minimum number of tokens for your model and platform (see Cache limitations)
• Confirm your breakpoint is on a block that stays identical across requests. Cache writes happen only at the breakpoint, and if that block changes (timestamps, per-request context, the incoming message), the prefix hash never matches. The lookback does not find stable content behind the breakpoint; it only finds entries that earlier requests wrote at their own breakpoints
• Verify that the keys in your tool_use content blocks have stable ordering as some languages (for example, Swift, Go) randomize key order during JSON conversion, breaking caches
• Use cache diagnostics to have the API compare consecutive requests and report which part of the prompt diverged

1-hour cache duration

If you find that 5 minutes is too short, Anthropic also offers a 1-hour cache duration at additional cost.

To use the extended cache, include ttl in the cache_control definition like this:

"cache_control": {
  "type": "ephemeral",
  "ttl": "1h"
}

The response includes detailed cache information like the following:

{
  "usage": {
    "input_tokens": 2048,
    "cache_read_input_tokens": 1800,
    "cache_creation_input_tokens": 248,
    "output_tokens": 503,

    "cache_creation": {
      "ephemeral_5m_input_tokens": 148,
      "ephemeral_1h_input_tokens": 100
    }
  }
}

Note that the current cache_creation_input_tokens field equals the sum of the values in the cache_creation object.

If you see ephemeral_5m_input_tokens writes you didn't request while using server tools such as web search, see Tool use with prompt caching.

When to use the 1-hour cache

If you have prompts that are used at a regular cadence (that is, system prompts that are used more frequently than every 5 minutes), continue to use the 5-minute cache, because this will continue to be refreshed at no additional charge.

The 1-hour cache is best used in the following scenarios:

• When you have prompts that are likely used less frequently than 5 minutes, but more frequently than every hour. For example, when an agentic side-agent will take longer than 5 minutes, or when storing a long chat conversation with a user and you generally expect that user may not respond in the next 5 minutes.
• When latency is important and your follow up prompts may be sent beyond 5 minutes.
• When you want to improve your rate limit utilization, because cache hits are not deducted against your rate limit.

Mixing different TTLs

You can use both 1-hour and 5-minute cache controls in the same request, but with an important constraint: Cache entries with longer TTL must appear before shorter TTLs (that is, a 1-hour cache entry must appear before any 5-minute cache entries).

When mixing TTLs, the API determines three billing locations in your prompt:

1. Position A: The token count at the highest cache hit (or 0 if no hits).
2. Position B: The token count at the highest 1-hour cache_control block after A (or equals A if none exist).
3. Position C: The token count at the last cache_control block.

You'll be charged for:

1. Cache read tokens for A.
2. 1-hour cache write tokens for (B - A).
3. 5-minute cache write tokens for (C - B).

Here are three examples. This depicts the input tokens of 3 requests, each of which has different cache hits and cache misses. Each has a different calculated pricing, shown in the colored boxes, as a result.

Pre-warming the cache

Cache pre-warming lets you load your system prompt or tool definitions into the prompt cache before a user triggers a real request. This eliminates the cache-miss latency penalty on the first user interaction, reducing time-to-first-token (TTFT) for latency-sensitive applications.

How it works

Set max_tokens: 0 in your request. The API reads your prompt into the model and writes the cache at any cache_control breakpoint, then returns immediately without generating any output. The response has an empty content array, stop_reason: "max_tokens", and a fully populated usage block.

Place the cache_control breakpoint on the last block that is shared with the follow-up request (typically your system prompt or tool definitions), not on the placeholder user message. Otherwise the cache entry is keyed to the placeholder and the follow-up request won't hit it. Use the same thinking configuration and output_config.effort as your follow-up requests too: those values are rendered into the prompt (see What invalidates the cache), so a pre-warm with a different configuration can write an entry your real traffic never hits. This means using an explicit cache breakpoint rather than automatic caching, since automatic caching places the breakpoint on the last block, which here is the placeholder. The placeholder user message can be any string with non-whitespace content (the examples here use "warmup"); its content is read into the model but never answered.

client = anthropic.Anthropic()

# Fire this before users arrive to warm the shared system-prompt cache.
prewarm = client.messages.create(
    model="claude-opus-5",
    max_tokens=0,
    system=[
        {
            "type": "text",
            "text": "You are an expert software engineer with deep knowledge of distributed systems...",
            "cache_control": {"type": "ephemeral"},
        }
    ],
    messages=[{"role": "user", "content": "warmup"}],
)
print(prewarm.stop_reason)  # "max_tokens"
print(prewarm.content)  # []
print(prewarm.usage)

The API returns an empty content array:

{
  "id": "msg_01XFDUDYJgAACzvnptvVoYEL",
  "type": "message",
  "role": "assistant",
  "content": [],
  "model": "claude-opus-5",
  "stop_reason": "max_tokens",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 8,
    "cache_creation_input_tokens": 5120,
    "cache_read_input_tokens": 0,
    "cache_creation": {
      "ephemeral_5m_input_tokens": 5120,
      "ephemeral_1h_input_tokens": 0
    },
    "iterations": [
      {
        "input_tokens": 8,
        "output_tokens": 0,
        "cache_read_input_tokens": 0,
        "cache_creation_input_tokens": 5120,
        "cache_creation": {
          "ephemeral_5m_input_tokens": 5120,
          "ephemeral_1h_input_tokens": 0
        },
        "type": "message"
      }
    ],
    "output_tokens": 0,
    "service_tier": "standard",
    "inference_geo": "global"
  }
}

Typical usage pattern

Fire a pre-warm request when your application starts (or on a scheduled interval), then send real user requests after the pre-warm completes:

client = anthropic.Anthropic()

SYSTEM_PROMPT = [
    {
        "type": "text",
        "text": "You are an expert software engineer with deep knowledge of distributed systems...",
        "cache_control": {"type": "ephemeral"},
    }
]


def prewarm_cache() -> None:
    """Call this at application startup or on a scheduled interval."""
    client.messages.create(
        model="claude-opus-5",
        max_tokens=0,
        system=SYSTEM_PROMPT,
        messages=[{"role": "user", "content": "warmup"}],
    )


def respond(user_message: str) -> anthropic.types.Message:
    """The real user request; benefits from a warm cache."""
    return client.messages.create(
        model="claude-opus-5",
        max_tokens=1024,
        system=SYSTEM_PROMPT,
        messages=[{"role": "user", "content": user_message}],
    )


# Warm the cache before any user traffic arrives.
prewarm_cache()

# Later, when the user submits a message, the system-prompt prefix is already cached.
response = respond("How do I implement a binary search tree?")
for block in response.content:
    if block.type == "text":
        print(block.text)

Keep in mind that the cache TTL still applies. For the default 5-minute cache, send a new pre-warm request at least every 5 minutes to keep the cache warm. For longer gaps between user requests, use the 1-hour cache duration instead.

Limitations

A max_tokens: 0 request is rejected with an invalid_request_error if any of the following are set, since each implies output that a zero-token budget cannot produce:

• stream: true
• Extended thinking (thinking.type: "enabled")
• Structured outputs (output_config.format)
• tool_choice of {"type": "tool", ...} or {"type": "any"}

max_tokens: 0 is also rejected inside a Message Batches request. Pre-warming targets time-to-first-token, which does not apply to batch processing, and a cache entry written during batch processing would likely expire before the follow-up request runs.

Replacing the max_tokens=1 workaround

Before max_tokens: 0 was available, some applications used max_tokens: 1 warm-up calls to achieve the same effect. The max_tokens: 0 approach is preferred: no output is produced, so there is no single-token reply to discard, no output tokens are billed, and the intent of the request is unambiguous.

Prompt caching examples

To help you get started with prompt caching, the prompt caching cookbook provides detailed examples and best practices.

The following code snippets showcase various prompt caching patterns. These examples demonstrate how to implement caching in different scenarios, helping you understand the practical applications of this feature:

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-5",
    max_tokens=1024,
    system=[
        {
            "type": "text",
            "text": "You are an AI assistant tasked with analyzing legal documents.",
        },
        {
            "type": "text",
            "text": "Here is the full text of a complex legal agreement: [Insert full text of a 50-page legal agreement here]",
            "cache_control": {"type": "ephemeral"},
        },
    ],
    messages=[
        {
            "role": "user",
            "content": "What are the key terms and conditions in this agreement?",
        }
    ],
)
print(response.usage.model_dump_json())

{
  "model": "claude-opus-5",
  "max_tokens": 1024,
  "tools": [
    {
      "name": "get_weather",
      "description": "Get the current weather in a given location",
      "input_schema": {
        "type": "object",
        "properties": { "location": { "type": "string" } },
        "required": ["location"]
      }
    },
    {
      "name": "get_time",
      "description": "Get the current time in a given time zone",
      "input_schema": {
        "type": "object",
        "properties": { "timezone": { "type": "string" } },
        "required": ["timezone"]
      },
      "cache_control": { "type": "ephemeral" }
    }
  ],
  "messages": [{ "role": "user", "content": "What is the weather and time in New York?" }]
}

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-5",
    max_tokens=1024,
    system=[
        {
            "type": "text",
            "text": "...long system prompt",
            "cache_control": {"type": "ephemeral"},
        }
    ],
    messages=[
        # ...long conversation so far
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "Hello, can you tell me more about the solar system?",
                }
            ],
        },
        {
            "role": "assistant",
            "content": "Certainly! The solar system is the collection of celestial bodies that orbit our Sun. It consists of eight planets, numerous moons, asteroids, comets, and other objects. The planets, in order from closest to farthest from the Sun, are: Mercury, Venus, Earth, Mars, Jupiter, Saturn, Uranus, and Neptune. Each planet has its own unique characteristics and features. Is there a specific aspect of the solar system you'd like to know more about?",
        },
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "Good to know."},
                {
                    "type": "text",
                    "text": "Tell me more about Mars.",
                    "cache_control": {"type": "ephemeral"},
                },
            ],
        },
    ],
)
print(response.usage.model_dump_json())

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-5",
    max_tokens=1024,
    tools=[
        {
            "name": "search_documents",
            "description": "Search through the knowledge base",
            "input_schema": {
                "type": "object",
                "properties": {
                    "query": {"type": "string", "description": "Search query"}
                },
                "required": ["query"],
            },
        },
        {
            "name": "get_document",
            "description": "Retrieve a specific document by ID",
            "input_schema": {
                "type": "object",
                "properties": {
                    "doc_id": {"type": "string", "description": "Document ID"}
                },
                "required": ["doc_id"],
            },
            "cache_control": {"type": "ephemeral"},
        },
    ],
    system=[
        {
            "type": "text",
            "text": "You are a helpful research assistant with access to a document knowledge base.\n\n# Instructions\n- Always search for relevant documents before answering\n- Provide citations for your sources\n- Be objective and accurate in your responses\n- If multiple documents contain relevant information, synthesize them\n- Acknowledge when information is not available in the knowledge base",
            "cache_control": {"type": "ephemeral"},
        },
        {
            "type": "text",
            "text": "# Knowledge Base Context\n\nHere are the relevant documents for this conversation:\n\n## Document 1: Solar System Overview\nThe solar system consists of the Sun and all objects that orbit it...\n\n## Document 2: Planetary Characteristics\nEach planet has unique features. Mercury is the smallest planet...\n\n## Document 3: Mars Exploration\nMars has been a target of exploration for decades...\n\n[Additional documents...]",
            "cache_control": {"type": "ephemeral"},
        },
    ],
    messages=[
        {
            "role": "user",
            "content": "Can you search for information about Mars rovers?",
        },
        {
            "role": "assistant",
            "content": [
                {
                    "type": "tool_use",
                    "id": "tool_1",
                    "name": "search_documents",
                    "input": {"query": "Mars rovers"},
                }
            ],
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "tool_result",
                    "tool_use_id": "tool_1",
                    "content": "Found 3 relevant documents: Document 3 (Mars Exploration), Document 7 (Rover Technology), Document 9 (Mission History)",
                }
            ],
        },
        {
            "role": "assistant",
            "content": [
                {
                    "type": "text",
                    "text": "I found 3 relevant documents about Mars rovers. Let me get more details from the Mars Exploration document.",
                }
            ],
        },
        {
            "role": "user",
            "content": [
                {
                    "type": "text",
                    "text": "Yes, please tell me about the Perseverance rover specifically.",
                    "cache_control": {"type": "ephemeral"},
                }
            ],
        },
    ],
)
print(response.usage.model_dump_json())

Data retention

Prompt caching (both automatic and explicit) is ZDR eligible. Anthropic does not store the raw text of your prompts or Claude's responses.

KV (key-value) cache representations and cryptographic hashes of cached content are held in memory only and are not stored at rest. Cached entries have a minimum lifetime of 5 minutes (standard) or 1 hour (extended), after which they are promptly, though not immediately, deleted. Cache entries are isolated between organizations and, on the Claude API, Claude Platform on AWS, and Microsoft Foundry, between workspaces within an organization.

For ZDR eligibility across all features, see API and data retention.

FAQ

total_input_tokens = cache_read_input_tokens + cache_creation_input_tokens + input_tokens

client.beta.prompt_caching.messages.create(**params)

client.messages.create(**params)

client.beta.promptCaching.messages.create(/* ... */);

client.messages.create(/* ... */);