# 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.¶
¶
<Note>¶
This feature is eligible for [Zero Data Retention (ZDR)](/docs/en/build-with-claude/api-and-data-retention). When your organization has a ZDR arrangement, data sent through this feature is not stored after the API response is returned.¶
</Note>¶
¶
There are two ways to enable prompt caching:¶
¶
- **[Automatic 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](#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:¶
¶
<CodeGroup>¶
¶
```bash Shell¶
curl https://api.anthropic.com/v1/messages \¶
-H "content-type: application/json" \¶
-H "x-api-key: $ANTHROPIC_API_KEY" \¶
-H "anthropic-version: 2023-06-01" \¶
-d '{¶
"model": "claude-opus-4-6",¶
"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."¶
}¶
]¶
}'¶
```¶
¶
```python Python hidelines={1..2}¶
import anthropic¶
¶
client = anthropic.Anthropic()¶
¶
response = client.messages.create(¶
model="claude-opus-4-6",¶
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())¶
```¶
¶
```typescript TypeScript hidelines={1..2}¶
import Anthropic from "@anthropic-ai/sdk";¶
¶
const client = new Anthropic();¶
¶
const response = await client.messages.create({¶
model: "claude-opus-4-6",¶
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'."¶
}¶
]¶
});¶
console.log(response.usage);¶
```¶
¶
```csharp C# hidelines={1..9,-2..}¶
using System;¶
using System.Threading.Tasks;¶
using Anthropic;¶
using Anthropic.Models.Messages;¶
¶
class Program¶
{¶
static async Task Main(string[] args)¶
{¶
AnthropicClient client = new();¶
¶
var parameters = new MessageCreateParams¶
{¶
Model = Model.ClaudeOpus4_6,¶
MaxTokens = 1024,¶
CacheControl = new CacheControlEphemeral(),¶
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 =¶
[¶
new()¶
{¶
Role = Role.User,¶
Content = "Analyze the major themes in 'Pride and Prejudice'."¶
}¶
]¶
};¶
¶
var message = await client.Messages.Create(parameters);¶
Console.WriteLine(message.Usage);¶
}¶
}¶
```¶
¶
```go Go hidelines={1..11,-1}¶
package main¶
¶
import (¶
"context"¶
"fmt"¶
"log"¶
¶
"github.com/anthropics/anthropic-sdk-go"¶
)¶
¶
func main() {¶
client := anthropic.NewClient()¶
¶
response, err := client.Messages.New(context.TODO(), anthropic.MessageNewParams{¶
Model: anthropic.ModelClaudeOpus4_6,¶
MaxTokens: 1024,¶
CacheControl: anthropic.NewCacheControlEphemeralParam(),¶
System: []anthropic.TextBlockParam{¶
{Text: "You are an AI assistant tasked with analyzing literary works. Your goal is to provide insightful commentary on themes, characters, and writing style."},¶
},¶
Messages: []anthropic.MessageParam{¶
anthropic.NewUserMessage(anthropic.NewTextBlock("Analyze the major themes in 'Pride and Prejudice'.")),¶
},¶
})¶
if err != nil {¶
log.Fatal(err)¶
}¶
fmt.Println(response.Usage)¶
}¶
```¶
¶
```java Java hidelines={1..2,4..10,-2..}¶
import com.anthropic.client.AnthropicClient;¶
import com.anthropic.client.okhttp.AnthropicOkHttpClient;¶
import com.anthropic.models.messages.CacheControlEphemeral;¶
import com.anthropic.models.messages.Message;¶
import com.anthropic.models.messages.MessageCreateParams;¶
import com.anthropic.models.messages.Model;¶
¶
public class PromptCachingExample {¶
¶
public static void main(String[] args) {¶
AnthropicClient client = AnthropicOkHttpClient.fromEnv();¶
¶
MessageCreateParams params = MessageCreateParams.builder()¶
.model(Model.CLAUDE_OPUS_4_6)¶
.maxTokens(1024)¶
.cacheControl(CacheControlEphemeral.builder().build())¶
.system("You are an AI assistant tasked with analyzing literary works. Your goal is to provide insightful commentary on themes, characters, and writing style.")¶
.addUserMessage("Analyze the major themes in 'Pride and Prejudice'.")¶
.build();¶
¶
Message message = client.messages().create(params);¶
System.out.println(message.usage());¶
}¶
}¶
```¶
¶
```php PHP hidelines={1..3,5}¶
<?php¶
¶
use Anthropic\Client;¶
use Anthropic\Messages\CacheControlEphemeral;¶
¶
$client = new Client(apiKey: getenv("ANTHROPIC_API_KEY"));¶
¶
$response = $client->messages->create(¶
maxTokens: 1024,¶
messages: [¶
['role' => 'user', 'content' => "Analyze the major themes in 'Pride and Prejudice'."]¶
],¶
model: 'claude-opus-4-6',¶
cacheControl: CacheControlEphemeral::with(),¶
system: "You are an AI assistant tasked with analyzing literary works. Your goal is to provide insightful commentary on themes, characters, and writing style.",¶
);¶
echo json_encode($response->usage);¶
```¶
¶
```ruby Ruby hidelines={1..2}¶
require "anthropic"¶
¶
client = Anthropic::Client.new¶
¶
response = client.messages.create(¶
model: "claude-opus-4-6",¶
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'."¶
}¶
]¶
)¶
puts response.usage¶
```¶
</CodeGroup>¶
¶
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.¶
¶
<Note>¶
If you find that 5 minutes is too short, Anthropic also offers a 1-hour cache duration [at additional cost](#pricing).¶
¶
For more information, see [1-hour cache duration](#1-hour-cache-duration).¶
</Note>¶
¶
<Tip>¶
**Prompt caching caches the full prefix**¶
¶
Prompt caching references the entire prompt - `tools`, `system`, and `messages` (in that order) up to and including the block designated with `cache_control`.¶
¶
</Tip>¶
¶
---¶
¶
## Pricing¶
¶
Prompt caching introduces a new pricing structure. The table below shows the price per million tokens for each supported model:¶
¶
| Model | Base Input Tokens | 5m Cache Writes | 1h Cache Writes | Cache Hits & Refreshes | Output Tokens |¶
|-------------------|-------------------|-----------------|-----------------|----------------------|---------------|¶
| Claude Opus 4.6 | $5 / MTok | $6.25 / MTok | $10 / MTok | $0.50 / MTok | $25 / MTok |¶
| Claude Opus 4.5 | $5 / MTok | $6.25 / MTok | $10 / MTok | $0.50 / MTok | $25 / MTok |¶
| Claude Opus 4.1 | $15 / MTok | $18.75 / MTok | $30 / MTok | $1.50 / MTok | $75 / MTok |¶
| Claude Opus 4 | $15 / MTok | $18.75 / MTok | $30 / MTok | $1.50 / MTok | $75 / MTok |¶
| Claude Sonnet 4.6 | $3 / MTok | $3.75 / MTok | $6 / MTok | $0.30 / MTok | $15 / MTok |¶
| Claude Sonnet 4.5 | $3 / MTok | $3.75 / MTok | $6 / MTok | $0.30 / MTok | $15 / MTok |¶
| Claude Sonnet 4 | $3 / MTok | $3.75 / MTok | $6 / MTok | $0.30 / MTok | $15 / MTok |¶
| Claude Sonnet 3.7 ([deprecated](/docs/en/about-claude/model-deprecations)) | $3 / MTok | $3.75 / MTok | $6 / MTok | $0.30 / MTok | $15 / MTok |¶
| Claude Haiku 4.5 | $1 / MTok | $1.25 / MTok | $2 / MTok | $0.10 / MTok | $5 / MTok |¶
| Claude Haiku 3.5 | $0.80 / MTok | $1 / MTok | $1.6 / MTok | $0.08 / MTok | $4 / MTok |¶
| Claude Opus 3 ([deprecated](/docs/en/about-claude/model-deprecations)) | $15 / MTok | $18.75 / MTok | $30 / MTok | $1.50 / MTok | $75 / MTok |¶
| Claude Haiku 3 | $0.25 / MTok | $0.30 / MTok | $0.50 / MTok | $0.03 / MTok | $1.25 / MTok |¶
¶
<Note>¶
The table above reflects the following pricing multipliers for prompt caching:¶
- 5-minute cache write tokens are 1.25 times the base input tokens price¶
- 1-hour cache write tokens are 2 times the base input tokens price¶
- Cache read tokens are 0.1 times the base input tokens price¶
¶
These multipliers stack with other pricing modifiers such as the Batch API discount, long context pricing, and data residency. See [pricing](/docs/en/about-claude/pricing) for full details.¶
</Note>¶
¶
---¶
¶
## Supported models¶
¶
Prompt caching (both automatic and explicit) is currently supported on:¶
- Claude Opus 4.6¶
- Claude Opus 4.5¶
- Claude Opus 4.1¶
- Claude Opus 4¶
- Claude Sonnet 4.6¶
- Claude Sonnet 4.5¶
- Claude Sonnet 4¶
- Claude Sonnet 3.7 ([deprecated](/docs/en/about-claude/model-deprecations))¶
- Claude Haiku 4.5¶
- Claude Haiku 3.5 ([deprecated](/docs/en/about-claude/model-deprecations))¶
- Claude Haiku 3¶
¶
---¶
¶
## 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.¶
¶
<CodeGroup>¶
¶
```bash Shell¶
curl https://api.anthropic.com/v1/messages \¶
-H "content-type: application/json" \¶
-H "x-api-key: $ANTHROPIC_API_KEY" \¶
-H "anthropic-version: 2023-06-01" \¶
-d '{¶
"model": "claude-opus-4-6",¶
"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?"}¶
]¶
}'¶
```¶
¶
```python Python hidelines={1..2}¶
import anthropic¶
¶
client = anthropic.Anthropic()¶
¶
response = client.messages.create(¶
model="claude-opus-4-6",¶
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())¶
```¶
¶
```typescript TypeScript hidelines={1..2}¶
import Anthropic from "@anthropic-ai/sdk";¶
¶
const client = new Anthropic();¶
¶
const response = await client.messages.create({¶
model: "claude-opus-4-6",¶
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?" }¶
]¶
});¶
console.log(response.usage);¶
```¶
¶
```csharp C# hidelines={1..9,-2..}¶
using System;¶
using System.Threading.Tasks;¶
using Anthropic;¶
using Anthropic.Models.Messages;¶
¶
class Program¶
{¶
static async Task Main(string[] args)¶
{¶
AnthropicClient client = new();¶
¶
var parameters = new MessageCreateParams¶
{¶
Model = Model.ClaudeOpus4_6,¶
MaxTokens = 1024,¶
CacheControl = new CacheControlEphemeral(),¶
System = "You are a helpful assistant that remembers our conversation.",¶
Messages =¶
[¶
new()¶
{¶
Role = Role.User,¶
Content = "My name is Alex. I work on machine learning."¶
},¶
new()¶
{¶
Role = Role.Assistant,¶
Content = "Nice to meet you, Alex! How can I help with your ML work today?"¶
},¶
new()¶
{¶
Role = Role.User,¶
Content = "What did I say I work on?"¶
}¶
]¶
};¶
¶
var message = await client.Messages.Create(parameters);¶
Console.WriteLine(message.Usage);¶
}¶
}¶
```¶
¶
```go Go hidelines={1..11,-1}¶
package main¶
¶
import (¶
"context"¶
"fmt"¶
"log"¶
¶
"github.com/anthropics/anthropic-sdk-go"¶
)¶
¶
func main() {¶
client := anthropic.NewClient()¶
¶
response, err := client.Messages.New(context.TODO(), anthropic.MessageNewParams{¶
Model: anthropic.ModelClaudeOpus4_6,¶
MaxTokens: 1024,¶
CacheControl: anthropic.NewCacheControlEphemeralParam(),¶
System: []anthropic.TextBlockParam{¶
{Text: "You are a helpful assistant that remembers our conversation."},¶
},¶
Messages: []anthropic.MessageParam{¶
anthropic.NewUserMessage(anthropic.NewTextBlock("My name is Alex. I work on machine learning.")),¶
anthropic.NewAssistantMessage(anthropic.NewTextBlock("Nice to meet you, Alex! How can I help with your ML work today?")),¶
anthropic.NewUserMessage(anthropic.NewTextBlock("What did I say I work on?")),¶
},¶
})¶
if err != nil {¶
log.Fatal(err)¶
}¶
fmt.Println(response.Usage)¶
}¶
```¶
¶
```java Java hidelines={1..2,4..10,-2..}¶
import com.anthropic.client.AnthropicClient;¶
import com.anthropic.client.okhttp.AnthropicOkHttpClient;¶
import com.anthropic.models.messages.CacheControlEphemeral;¶
import com.anthropic.models.messages.Message;¶
import com.anthropic.models.messages.MessageCreateParams;¶
import com.anthropic.models.messages.Model;¶
¶
public class AutomaticCachingExample {¶
¶
public static void main(String[] args) {¶
AnthropicClient client = AnthropicOkHttpClient.fromEnv();¶
¶
MessageCreateParams params = MessageCreateParams.builder()¶
.model(Model.CLAUDE_OPUS_4_6)¶
.maxTokens(1024)¶
.cacheControl(CacheControlEphemeral.builder().build())¶
.system("You are a helpful assistant that remembers our conversation.")¶
.addUserMessage("My name is Alex. I work on machine learning.")¶
.addAssistantMessage("Nice to meet you, Alex! How can I help with your ML work today?")¶
.addUserMessage("What did I say I work on?")¶
.build();¶
¶
Message message = client.messages().create(params);¶
System.out.println(message.usage());¶
}¶
}¶
```¶
¶
```php PHP hidelines={1..3,5}¶
<?php¶
¶
use Anthropic\Client;¶
use Anthropic\Messages\CacheControlEphemeral;¶
¶
$client = new Client(apiKey: getenv("ANTHROPIC_API_KEY"));¶
¶
$response = $client->messages->create(¶
maxTokens: 1024,¶
messages: [¶
['role' => 'user', 'content' => 'My name is Alex. I work on machine learning.'],¶
['role' => 'assistant', 'content' => 'Nice to meet you, Alex! How can I help with your ML work today?'],¶
['role' => 'user', 'content' => 'What did I say I work on?'],¶
],¶
model: 'claude-opus-4-6',¶
cacheControl: CacheControlEphemeral::with(),¶
system: 'You are a helpful assistant that remembers our conversation.',¶
);¶
echo json_encode($response->usage);¶
```¶
¶
```ruby Ruby hidelines={1..2}¶
require "anthropic"¶
¶
client = Anthropic::Client.new¶
¶
response = client.messages.create(¶
model: "claude-opus-4-6",¶
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?"}¶
]¶
)¶
puts response.usage¶
```¶
</CodeGroup>¶
¶
### 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.¶
¶
| Request | Content | Cache behavior |¶
|---------|---------|----------------|¶
| Request 1 | System <br/> + User(1) + Asst(1) <br/> + **User(2)** ◀ cache | Everything written to cache |¶
| Request 2 | System <br/> + User(1) + Asst(1) <br/> + User(2) + Asst(2) <br/> + **User(3)** ◀ cache | System through User(2) read from cache; <br/> Asst(2) + User(3) written to cache |¶
| Request 3 | System <br/> + User(1) + Asst(1) <br/> + User(2) + Asst(2) <br/> + User(3) + Asst(3) <br/> + **User(4)** ◀ cache | System through User(3) read from cache; <br/> Asst(3) + User(4) written to 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:¶
¶
```json¶
{ "cache_control": { "type": "ephemeral", "ttl": "1h" } }¶
```¶
¶
### Combining with block-level caching¶
¶
Automatic caching is compatible with [explicit cache breakpoints](#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 explicit breakpoints to cache your system prompt and tools independently, while automatic caching handles the conversation:¶
¶
```json¶
{¶
"model": "claude-opus-4-6",¶
"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.¶
¶
<Note>¶
Automatic caching is available on the Claude API and Azure AI Foundry (preview). Support for Amazon Bedrock and Google Vertex AI is coming later.¶
</Note>¶
¶
---¶
¶
## 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).¶
¶
**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](#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¶
¶
<Note>¶
**Important limitation:** The lookback can only find entries that earlier requests already wrote. If a growing conversation pushes your breakpoint 20 or more blocks past the last write, the lookback window misses it. Add a second breakpoint closer to that position from the start so a write accumulates there before you need it.¶
</Note>¶
¶
### 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)¶
- **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 simply give you control over what sections can be cached independently.¶
¶
---¶
¶
## Caching strategies and considerations¶
¶
### Cache limitations¶
The minimum cacheable prompt length is:¶
- 4096 tokens for Claude Opus 4.6, Claude Opus 4.5¶
- 2048 tokens for Claude Sonnet 4.6¶
- 1024 tokens for Claude Sonnet 4.5, Claude Opus 4.1, Claude Opus 4, Claude Sonnet 4, and Claude Sonnet 3.7 ([deprecated](/docs/en/about-claude/model-deprecations))¶
- 4096 tokens for Claude Haiku 4.5¶
- 2048 tokens for Claude Haiku 3.5 ([deprecated](/docs/en/about-claude/model-deprecations)) and Claude Haiku 3¶
¶
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](/docs/en/build-with-claude/prompt-caching#tracking-cache-performance): 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 the model you are using, 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](/docs/en/build-with-claude/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](#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.¶
¶
| What changes | Tools cache | System cache | Messages cache | Impact |¶
|------------|------------------|---------------|----------------|-------------|¶
| **Tool definitions** | ✘ | ✘ | ✘ | Modifying tool definitions (names, descriptions, parameters) invalidates the entire cache |¶
| **Web search toggle** | ✓ | ✘ | ✘ | Enabling/disabling web search modifies the system prompt |¶
| **Citations toggle** | ✓ | ✘ | ✘ | Enabling/disabling citations modifies the system prompt |¶
| **Speed setting** | ✓ | ✘ | ✘ | Switching between [`speed: "fast"` and standard speed](/docs/en/build-with-claude/fast-mode) invalidates system and message caches |¶
| **Tool choice** | ✓ | ✓ | ✘ | Changes to `tool_choice` parameter only affect message blocks |¶
| **Images** | ✓ | ✓ | ✘ | Adding/removing images anywhere in the prompt affects message blocks |¶
| **Thinking parameters** | ✓ | ✓ | ✘ | Changes to extended thinking settings (enable/disable, budget) affect message blocks |¶
| **Non-tool results passed to extended thinking requests** | ✓ | ✓ | ✘ | When non-tool results are passed in requests while extended thinking is enabled, all previously-cached thinking blocks are stripped from context, and any messages in context that follow those thinking blocks are removed from the cache. For more details, see [Caching with thinking blocks](#caching-with-thinking-blocks). |¶
¶
### Tracking cache performance¶
¶
Monitor cache performance using these API response fields, within `usage` in the response (or `message_start` event if [streaming](/docs/en/build-with-claude/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).¶
¶
<Note>¶
**Understanding the token breakdown**¶
¶
The `input_tokens` field represents only the tokens that come **after the last cache breakpoint** in your request - not all the input tokens you sent.¶
¶
To calculate total input tokens:¶
```text¶
total_input_tokens = cache_read_input_tokens + cache_creation_input_tokens + input_tokens¶
```¶
¶
**Spatial explanation:**¶
- `cache_read_input_tokens` = tokens before breakpoint already cached (reads)¶
- `cache_creation_input_tokens` = tokens before breakpoint being cached now (writes)¶
- `input_tokens` = tokens after your last breakpoint (not eligible for cache)¶
¶
**Example:** If you have a request with 100,000 tokens of cached content (read from cache), 0 tokens of new content being cached, and 50 tokens in your user message (after the cache breakpoint):¶
- `cache_read_input_tokens`: 100,000¶
- `cache_creation_input_tokens`: 0¶
- `input_tokens`: 50¶
- **Total input tokens processed**: 100,050 tokens¶
¶
This is important for understanding both costs and rate limits, as `input_tokens` will typically be much smaller than your total input when using caching effectively.¶
</Note>¶
¶
### Caching with thinking blocks¶
¶
When using [extended thinking](/docs/en/build-with-claude/extended-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¶
- Cache gets invalidated when non-tool-result user content is added, causing all previous thinking blocks to be stripped¶
- This caching behavior occurs even without explicit `cache_control` markers¶
¶
For more details on cache invalidation, see [What invalidates the cache](#what-invalidates-the-cache).¶
¶
**Example with tool use**:¶
```text¶
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]¶
# Non-tool-result user block causes all thinking blocks to be ignored¶
# This request is processed as if thinking blocks were never present¶
```¶
¶
When a non-tool-result user block is included, it designates a new assistant loop and all previous thinking blocks are removed from context.¶
¶
For more detailed information, see the [extended thinking documentation](/docs/en/build-with-claude/extended-thinking#understanding-thinking-block-caching-behavior).¶
¶
### Cache storage and sharing¶
¶
<Warning>¶
Starting February 5, 2026, prompt caching will use workspace-level isolation instead of organization-level isolation. Caches will be isolated per workspace, ensuring data separation between workspaces within the same organization. This change applies to the Claude API and Azure AI Foundry (preview); Amazon Bedrock and Google Vertex AI will maintain organization-level cache isolation. If you use multiple workspaces, review your caching strategy to account for this change.¶
</Warning>¶
¶
- **Organization Isolation**: Caches are isolated between organizations. Different organizations never share caches, even if they use identical prompts.¶
¶
- **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 will be identical to what you would get if prompt caching was not used.¶
¶
### Best practices for effective caching¶
¶
To optimize prompt caching performance:¶
¶
- Start with [automatic caching](#automatic-caching) for multi-turn conversations. It handles breakpoint management automatically.¶
- Use [explicit block-level breakpoints](#explicit-cache-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` and image usage remain consistent between calls¶
- Validate that you are caching at least the minimum number of tokens for the model you are using (see [Cache limitations](#cache-limitations)). Length-based caching failures are silent: the request succeeds but both `cache_creation_input_tokens` and `cache_read_input_tokens` will be 0¶
- 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¶
¶
<Note>¶
Changes to `tool_choice` or the presence/absence of images anywhere in the prompt will invalidate the cache, requiring a new cache entry to be created. For more details on cache invalidation, see [What invalidates the cache](#what-invalidates-the-cache).¶
</Note>¶
¶
---¶
## 1-hour cache duration¶
¶
If you find that 5 minutes is too short, Anthropic also offers a 1-hour cache duration [at additional cost](#pricing).¶
¶
To use the extended cache, include `ttl` in the `cache_control` definition like this:¶
```json hidelines={1,-1}¶
{¶
"cache_control": {¶
"type": "ephemeral",¶
"ttl": "1h"¶
}¶
}¶
```¶
¶
The response will include detailed cache information like the following:¶
```json¶
{¶
"usage": {¶
"input_tokens": 2048,¶
"cache_read_input_tokens": 1800,¶
"cache_creation_input_tokens": 248,¶
"output_tokens": 503,¶
¶
"cache_creation": {¶
"ephemeral_5m_input_tokens": 456,¶
"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.¶
¶
### 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, since 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, since cache hits are not deducted against your rate limit.¶
¶
<Note>¶
The 5-minute and 1-hour cache behave the same with respect to latency. You will generally see improved time-to-first-token for long documents.¶
</Note>¶
¶
### 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.¶
¶
<Note>¶
If `B` and/or `C` are larger than `A`, they will necessarily be cache misses, because `A` is the highest cache hit.¶
</Note>¶
¶
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 3 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.¶
¶
¶
---¶
## Prompt caching examples¶
¶
To help you get started with prompt caching, the [prompt caching cookbook](https://platform.claude.com/cookbook/misc-prompt-caching) 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:¶
¶
<section title="Large context caching example">¶
¶
<CodeGroup>¶
```bash Shell¶
curl https://api.anthropic.com/v1/messages \¶
--header "x-api-key: $ANTHROPIC_API_KEY" \¶
--header "anthropic-version: 2023-06-01" \¶
--header "content-type: application/json" \¶
--data \¶
'{¶
"model": "claude-opus-4-6",¶
"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?"¶
}¶
]¶
}'¶
```¶
¶
```python Python hidelines={1..2}¶
import anthropic¶
¶
client = anthropic.Anthropic()¶
¶
response = client.messages.create(¶
model="claude-opus-4-6",¶
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.model_dump_json())¶
```¶
¶
```typescript TypeScript hidelines={1..2}¶
import Anthropic from "@anthropic-ai/sdk";¶
¶
const client = new Anthropic();¶
¶
const response = await client.messages.create({¶
model: "claude-opus-4-6",¶
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?"¶
}¶
]¶
});¶
console.log(response);¶
```¶
¶
```csharp C# hidelines={1..10,-2..}¶
using System;¶
using System.Threading.Tasks;¶
using System.Collections.Generic;¶
using Anthropic;¶
using Anthropic.Models.Messages;¶
¶
public class Program¶
{¶
public static async Task Main(string[] args)¶
{¶
AnthropicClient client = new()¶
{¶
ApiKey = Environment.GetEnvironmentVariable("ANTHROPIC_API_KEY")¶
};¶
¶
var parameters = new MessageCreateParams¶
{¶
Model = Model.ClaudeOpus4_6,¶
MaxTokens = 1024,¶
System = new MessageCreateParamsSystem(new List<TextBlockParam>¶
{¶
new TextBlockParam()¶
{¶
Text = "You are an AI assistant tasked with analyzing legal documents.",¶
},¶
new TextBlockParam()¶
{¶
Text = "Here is the full text of a complex legal agreement: [Insert full text of a 50-page legal agreement here]",¶
CacheControl = new CacheControlEphemeral(),¶
},¶
}),¶
Messages =¶
[¶
new()¶
{¶
Role = Role.User,¶
Content = "What are the key terms and conditions in this agreement?"¶
}¶
]¶
};¶
¶
var message = await client.Messages.Create(parameters);¶
Console.WriteLine(message);¶
}¶
}¶
```¶
¶
```go Go hidelines={1..11,-1}¶
package main¶
¶
import (¶
"context"¶
"fmt"¶
"log"¶
¶
"github.com/anthropics/anthropic-sdk-go"¶
)¶
¶
func main() {¶
client := anthropic.NewClient()¶
¶
response, err := client.Messages.New(context.TODO(), anthropic.MessageNewParams{¶
Model: anthropic.ModelClaudeOpus4_6,¶
MaxTokens: 1024,¶
System: []anthropic.TextBlockParam{¶
{¶
Text: "You are an AI assistant tasked with analyzing legal documents.",¶
},¶
{¶
Text: "Here is the full text of a complex legal agreement: [Insert full text of a 50-page legal agreement here]",¶
CacheControl: anthropic.NewCacheControlEphemeralParam(),¶
},¶
},¶
Messages: []anthropic.MessageParam{¶
anthropic.NewUserMessage(anthropic.NewTextBlock("What are the key terms and conditions in this agreement?")),¶
},¶
})¶
if err != nil {¶
log.Fatal(err)¶
}¶
fmt.Printf("%+v\n", response)¶
}¶
```¶
¶
```java Java hidelines={1..2,4..12,-2..}¶
import com.anthropic.client.AnthropicClient;¶
import com.anthropic.client.okhttp.AnthropicOkHttpClient;¶
import com.anthropic.models.messages.CacheControlEphemeral;¶
import com.anthropic.models.messages.Message;¶
import com.anthropic.models.messages.MessageCreateParams;¶
import com.anthropic.models.messages.Model;¶
import com.anthropic.models.messages.TextBlockParam;¶
import java.util.List;¶
¶
public class LegalDocumentAnalysisExample {¶
¶
public static void main(String[] args) {¶
AnthropicClient client = AnthropicOkHttpClient.fromEnv();¶
¶
MessageCreateParams params = MessageCreateParams.builder()¶
.model(Model.CLAUDE_OPUS_4_6)¶
.maxTokens(1024)¶
.systemOfTextBlockParams(¶
List.of(¶
TextBlockParam.builder()¶
.text("You are an AI assistant tasked with analyzing legal documents.")¶
.build(),¶
TextBlockParam.builder()¶
.text(¶
"Here is the full text of a complex legal agreement: [Insert full text of a 50-page legal agreement here]"¶
)¶
.cacheControl(CacheControlEphemeral.builder().build())¶
.build()¶
)¶
)¶
.addUserMessage("What are the key terms and conditions in this agreement?")¶
.build();¶
¶
Message message = client.messages().create(params);¶
System.out.println(message);¶
}¶
}¶
```¶
¶
```php PHP hidelines={1..4}¶
<?php¶
¶
use Anthropic\Client;¶
¶
$client = new Client(apiKey: getenv("ANTHROPIC_API_KEY"));¶
¶
$message = $client->messages->create(¶
maxTokens: 1024,¶
messages: [¶
[¶
'role' => 'user',¶
'content' => 'What are the key terms and conditions in this agreement?'¶
]¶
],¶
model: 'claude-opus-4-6',¶
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']¶
]¶
],¶
);¶
¶
echo $message->content[0]->text;¶
```¶
¶
```ruby Ruby nocheck hidelines={1..2}¶
require "anthropic"¶
¶
client = Anthropic::Client.new¶
¶
message = client.messages.create(¶
model: "claude-opus-4-6",¶
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?"¶
}¶
]¶
)¶
puts message¶
```¶
</CodeGroup>¶
This example demonstrates basic prompt caching usage, caching the full text of the legal agreement as a prefix while keeping the user instruction uncached.¶
¶
For the first request:¶
- `input_tokens`: Number of tokens in the user message only¶
- `cache_creation_input_tokens`: Number of tokens in the entire system message, including the legal document¶
- `cache_read_input_tokens`: 0 (no cache hit on first request)¶
¶
For subsequent requests within the cache lifetime:¶
- `input_tokens`: Number of tokens in the user message only¶
- `cache_creation_input_tokens`: 0 (no new cache creation)¶
- `cache_read_input_tokens`: Number of tokens in the entire cached system message¶
¶
</section>¶
¶
<section title="Caching tool definitions">¶
¶
Tool definitions can be cached by placing `cache_control` on the last tool in your `tools` array. All tools defined before and including that tool are cached as a single prefix.¶
¶
```json¶
{¶
"model": "claude-opus-4-6",¶
"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?" }]¶
}¶
```¶
¶
On the first request, `cache_creation_input_tokens` reflects the token count of all tool definitions. On subsequent requests within the cache lifetime, those tokens appear under `cache_read_input_tokens` instead.¶
¶
For detailed interaction between tool definitions, `defer_loading`, and cache invalidation, see [Tool use with prompt caching](/docs/en/agents-and-tools/tool-use/tool-use-with-prompt-caching).¶
¶
</section>¶
¶
<section title="Continuing a multi-turn conversation">¶
¶
<CodeGroup>¶
¶
```bash Shell¶
curl https://api.anthropic.com/v1/messages \¶
--header "x-api-key: $ANTHROPIC_API_KEY" \¶
--header "anthropic-version: 2023-06-01" \¶
--header "content-type: application/json" \¶
--data \¶
'{¶
"model": "claude-opus-4-6",¶
"max_tokens": 1024,¶
"system": [¶
{¶
"type": "text",¶
"text": "...long system prompt",¶
"cache_control": {"type": "ephemeral"}¶
}¶
],¶
"messages": [¶
{¶
"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 would 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"}¶
}¶
]¶
}¶
]¶
}'¶
```¶
¶
```python Python hidelines={1..2}¶
import anthropic¶
¶
client = anthropic.Anthropic()¶
¶
response = client.messages.create(¶
model="claude-opus-4-6",¶
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.model_dump_json())¶
```¶
¶
```typescript TypeScript hidelines={1..2}¶
import Anthropic from "@anthropic-ai/sdk";¶
¶
const client = new Anthropic();¶
¶
const response = await client.messages.create({¶
model: "claude-opus-4-6",¶
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" }¶
}¶
]¶
}¶
]¶
});¶
console.log(response);¶
```¶
¶
```csharp C# hidelines={1..6}¶
using Anthropic;¶
using Anthropic.Models.Messages;¶
using System.Collections.Generic;¶
¶
AnthropicClient client = new();¶
¶
var parameters = new MessageCreateParams¶
{¶
Model = Model.ClaudeOpus4_6,¶
MaxTokens = 1024,¶
System = new MessageCreateParamsSystem(new List<TextBlockParam>¶
{¶
new TextBlockParam()¶
{¶
Text = "...long system prompt",¶
CacheControl = new CacheControlEphemeral(),¶
},¶
}),¶
Messages =¶
[¶
new()¶
{¶
Role = Role.User,¶
Content = new MessageParamContent(new List<ContentBlockParam>¶
{¶
new ContentBlockParam(new TextBlockParam("Hello, can you tell me more about the solar system?")),¶
}),¶
},¶
new()¶
{¶
Role = 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 would like to know more about?"¶
},¶
new()¶
{¶
Role = Role.User,¶
Content = new MessageParamContent(new List<ContentBlockParam>¶
{¶
new ContentBlockParam(new TextBlockParam("Good to know.")),¶
new ContentBlockParam(new TextBlockParam()¶
{¶
Text = "Tell me more about Mars.",¶
CacheControl = new CacheControlEphemeral(),¶
}),¶
})¶
}¶
]¶
};¶
¶
var message = await client.Messages.Create(parameters);¶
Console.WriteLine(message);¶
```¶
¶
```go Go hidelines={1..11,-1}¶
package main¶
¶
import (¶
"context"¶
"fmt"¶
"log"¶
¶
"github.com/anthropics/anthropic-sdk-go"¶
)¶
¶
func main() {¶
client := anthropic.NewClient()¶
¶
response, err := client.Messages.New(context.TODO(), anthropic.MessageNewParams{¶
Model: anthropic.ModelClaudeOpus4_6,¶
MaxTokens: 1024,¶
System: []anthropic.TextBlockParam{¶
{¶
Text: "...long system prompt",¶
CacheControl: anthropic.NewCacheControlEphemeralParam(),¶
},¶
},¶
Messages: []anthropic.MessageParam{¶
anthropic.NewUserMessage(anthropic.NewTextBlock("Hello, can you tell me more about the solar system?")),¶
anthropic.NewAssistantMessage(anthropic.NewTextBlock("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 would like to know more about?")),¶
{¶
Role: anthropic.MessageParamRoleUser,¶
Content: []anthropic.ContentBlockParamUnion{¶
anthropic.NewTextBlock("Good to know."),¶
{OfText: &anthropic.TextBlockParam{¶
Text: "Tell me more about Mars.",¶
CacheControl: anthropic.NewCacheControlEphemeralParam(),¶
}},¶
},¶
},¶
},¶
})¶
if err != nil {¶
log.Fatal(err)¶
}¶
fmt.Println(response)¶
}¶
```¶
¶
```java Java hidelines={1..2,4..13,-2..}¶
import com.anthropic.client.AnthropicClient;¶
import com.anthropic.client.okhttp.AnthropicOkHttpClient;¶
import com.anthropic.models.messages.CacheControlEphemeral;¶
import com.anthropic.models.messages.ContentBlockParam;¶
import com.anthropic.models.messages.Message;¶
import com.anthropic.models.messages.MessageCreateParams;¶
import com.anthropic.models.messages.Model;¶
import com.anthropic.models.messages.TextBlockParam;¶
import java.util.List;¶
¶
public class ConversationWithCacheControlExample {¶
¶
public static void main(String[] args) {¶
AnthropicClient client = AnthropicOkHttpClient.fromEnv();¶
¶
// Create ephemeral system prompt¶
TextBlockParam systemPrompt = TextBlockParam.builder()¶
.text("...long system prompt")¶
.cacheControl(CacheControlEphemeral.builder().build())¶
.build();¶
¶
// Create message params¶
MessageCreateParams params = MessageCreateParams.builder()¶
.model(Model.CLAUDE_OPUS_4_6)¶
.maxTokens(1024)¶
.systemOfTextBlockParams(List.of(systemPrompt))¶
// First user message (without cache control)¶
.addUserMessage("Hello, can you tell me more about the solar system?")¶
// Assistant response¶
.addAssistantMessage(¶
"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 would like to know more about?"¶
)¶
// Second user message (with cache control)¶
.addUserMessageOfBlockParams(¶
List.of(¶
ContentBlockParam.ofText(TextBlockParam.builder().text("Good to know.").build()),¶
ContentBlockParam.ofText(¶
TextBlockParam.builder()¶
.text("Tell me more about Mars.")¶
.cacheControl(CacheControlEphemeral.builder().build())¶
.build()¶
)¶
)¶
)¶
.build();¶
¶
Message message = client.messages().create(params);¶
System.out.println(message);¶
}¶
}¶
```¶
¶
```php PHP hidelines={1..4}¶
<?php¶
¶
use Anthropic\Client;¶
¶
$client = new Client(apiKey: getenv("ANTHROPIC_API_KEY"));¶
¶
$message = $client->messages->create(¶
maxTokens: 1024,¶
messages: [¶
[¶
'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 would 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']¶
]¶
]¶
]¶
],¶
model: 'claude-opus-4-6',¶
system: [¶
[¶
'type' => 'text',¶
'text' => '...long system prompt',¶
'cache_control' => ['type' => 'ephemeral']¶
]¶
],¶
);¶
¶
echo $message->content[0]->text;¶
```¶
¶
```ruby Ruby nocheck hidelines={1..2}¶
require "anthropic"¶
¶
client = Anthropic::Client.new¶
¶
message = client.messages.create(¶
model: "claude-opus-4-6",¶
max_tokens: 1024,¶
system: [¶
{¶
type: "text",¶
text: "...long system prompt",¶
cache_control: { type: "ephemeral" }¶
}¶
],¶
messages: [¶
{¶
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 would 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" }¶
}¶
]¶
}¶
]¶
)¶
puts message¶
```¶
</CodeGroup>¶
¶
This example demonstrates how to use prompt caching in a multi-turn conversation.¶
¶
During each turn, the final block of the final message is marked with `cache_control` so the conversation can be incrementally cached. The system will automatically lookup and use the longest previously cached sequence of blocks for follow-up messages. That is, blocks that were previously marked with a `cache_control` block are later not marked with this, but they will still be considered a cache hit (and also a cache refresh!) if they are hit within 5 minutes.¶
¶
In addition, note that the `cache_control` parameter is placed on the system message. This is to ensure that if this gets evicted from the cache (after not being used for more than 5 minutes), it will get added back to the cache on the next request.¶
¶
This approach is useful for maintaining context in ongoing conversations without repeatedly processing the same information.¶
¶
When this is set up properly, you should see the following in the usage response of each request:¶
- `input_tokens`: Number of tokens in the new user message (will be minimal)¶
- `cache_creation_input_tokens`: Number of tokens in the new assistant and user turns¶
- `cache_read_input_tokens`: Number of tokens in the conversation up to the previous turn¶
¶
</section>¶
¶
<section title="Putting it all together: Multiple cache breakpoints">¶
¶
<CodeGroup>¶
¶
```bash Shell¶
curl https://api.anthropic.com/v1/messages \¶
--header "x-api-key: $ANTHROPIC_API_KEY" \¶
--header "anthropic-version: 2023-06-01" \¶
--header "content-type: application/json" \¶
--data \¶
'{¶
"model": "claude-opus-4-6",¶
"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"}¶
}¶
]¶
}¶
]¶
}'¶
```¶
¶
```python Python hidelines={1..2}¶
import anthropic¶
¶
client = anthropic.Anthropic()¶
¶
response = client.messages.create(¶
model="claude-opus-4-6",¶
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.model_dump_json())¶
```¶
¶
```typescript TypeScript hidelines={1..2}¶
import Anthropic from "@anthropic-ai/sdk";¶
¶
const client = new Anthropic();¶
¶
const response = await client.messages.create({¶
model: "claude-opus-4-6",¶
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" }¶
}¶
]¶
}¶
]¶
});¶
console.log(response);¶
```¶
¶
```csharp C# hidelines={1..11,-2..}¶
using System;¶
using System.Collections.Generic;¶
using System.Text.Json;¶
using System.Threading.Tasks;¶
using Anthropic;¶
using Anthropic.Models.Messages;¶
¶
public class Program¶
{¶
public static async Task Main(string[] args)¶
{¶
AnthropicClient client = new()¶
{¶
ApiKey = Environment.GetEnvironmentVariable("ANTHROPIC_API_KEY")¶
};¶
¶
var parameters = new MessageCreateParams¶
{¶
Model = Model.ClaudeOpus4_6,¶
MaxTokens = 1024,¶
Tools =¶
[¶
new ToolUnion(new Tool()¶
{¶
Name = "search_documents",¶
Description = "Search through the knowledge base",¶
InputSchema = new InputSchema()¶
{¶
Properties = new Dictionary<string, JsonElement>¶
{¶
["query"] = JsonSerializer.SerializeToElement(new { type = "string", description = "Search query" }),¶
},¶
Required = ["query"],¶
},¶
}),¶
new ToolUnion(new Tool()¶
{¶
Name = "get_document",¶
Description = "Retrieve a specific document by ID",¶
InputSchema = new InputSchema()¶
{¶
Properties = new Dictionary<string, JsonElement>¶
{¶
["doc_id"] = JsonSerializer.SerializeToElement(new { type = "string", description = "Document ID" }),¶
},¶
Required = ["doc_id"],¶
},¶
CacheControl = new CacheControlEphemeral(),¶
}),¶
],¶
System = new MessageCreateParamsSystem(new List<TextBlockParam>¶
{¶
new TextBlockParam()¶
{¶
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",¶
CacheControl = new CacheControlEphemeral(),¶
},¶
new TextBlockParam()¶
{¶
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...]",¶
CacheControl = new CacheControlEphemeral(),¶
},¶
}),¶
Messages =¶
[¶
new() { Role = Role.User, Content = "Can you search for information about Mars rovers?" },¶
new()¶
{¶
Role = Role.Assistant,¶
Content = new MessageParamContent(new List<ContentBlockParam>¶
{¶
new ContentBlockParam(new ToolUseBlockParam()¶
{¶
ID = "tool_1",¶
Name = "search_documents",¶
Input = new Dictionary<string, JsonElement>¶
{¶
["query"] = JsonSerializer.SerializeToElement("Mars rovers"),¶
},¶
}),¶
}),¶
},¶
new()¶
{¶
Role = Role.User,¶
Content = new MessageParamContent(new List<ContentBlockParam>¶
{¶
new ContentBlockParam(new ToolResultBlockParam()¶
{¶
ToolUseID = "tool_1",¶
Content = "Found 3 relevant documents: Document 3 (Mars Exploration), Document 7 (Rover Technology), Document 9 (Mission History)",¶
}),¶
}),¶
},¶
new()¶
{¶
Role = Role.Assistant,¶
Content = "I found 3 relevant documents about Mars rovers. Let me get more details from the Mars Exploration document.",¶
},¶
new()¶
{¶
Role = Role.User,¶
Content = new MessageParamContent(new List<ContentBlockParam>¶
{¶
new ContentBlockParam(new TextBlockParam()¶
{¶
Text = "Yes, please tell me about the Perseverance rover specifically.",¶
CacheControl = new CacheControlEphemeral(),¶
}),¶
}),¶
},¶
]¶
};¶
¶
var message = await client.Messages.Create(parameters);¶
Console.WriteLine(message);¶
}¶
}¶
```¶
¶
```go Go hidelines={1..11,-1}¶
package main¶
¶
import (¶
"context"¶
"fmt"¶
"log"¶
¶
"github.com/anthropics/anthropic-sdk-go"¶
)¶
¶
func main() {¶
client := anthropic.NewClient()¶
¶
response, err := client.Messages.New(context.TODO(), anthropic.MessageNewParams{¶
Model: anthropic.ModelClaudeOpus4_6,¶
MaxTokens: 1024,¶
Tools: []anthropic.ToolUnionParam{¶
{OfTool: &anthropic.ToolParam{¶
Name: "search_documents",¶
Description: anthropic.String("Search through the knowledge base"),¶
InputSchema: anthropic.ToolInputSchemaParam{¶
Properties: map[string]any{¶
"query": map[string]any{¶
"type": "string",¶
"description": "Search query",¶
},¶
},¶
Required: []string{"query"},¶
},¶
}},¶
{OfTool: &anthropic.ToolParam{¶
Name: "get_document",¶
Description: anthropic.String("Retrieve a specific document by ID"),¶
InputSchema: anthropic.ToolInputSchemaParam{¶
Properties: map[string]any{¶
"doc_id": map[string]any{¶
"type": "string",¶
"description": "Document ID",¶
},¶
},¶
Required: []string{"doc_id"},¶
},¶
CacheControl: anthropic.NewCacheControlEphemeralParam(),¶
}},¶
},¶
System: []anthropic.TextBlockParam{¶
{¶
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",¶
CacheControl: anthropic.NewCacheControlEphemeralParam(),¶
},¶
{¶
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...]",¶
CacheControl: anthropic.NewCacheControlEphemeralParam(),¶
},¶
},¶
Messages: []anthropic.MessageParam{¶
anthropic.NewUserMessage(anthropic.NewTextBlock("Can you search for information about Mars rovers?")),¶
anthropic.NewAssistantMessage(anthropic.NewToolUseBlock(¶
"tool_1",¶
map[string]any{"query": "Mars rovers"},¶
"search_documents",¶
)),¶
anthropic.NewUserMessage(anthropic.NewToolResultBlock(¶
"tool_1",¶
"Found 3 relevant documents: Document 3 (Mars Exploration), Document 7 (Rover Technology), Document 9 (Mission History)",¶
false,¶
)),¶
anthropic.NewAssistantMessage(anthropic.NewTextBlock("I found 3 relevant documents about Mars rovers. Let me get more details from the Mars Exploration document.")),¶
{¶
Role: anthropic.MessageParamRoleUser,¶
Content: []anthropic.ContentBlockParamUnion{¶
{OfText: &anthropic.TextBlockParam{¶
Text: "Yes, please tell me about the Perseverance rover specifically.",¶
CacheControl: anthropic.NewCacheControlEphemeralParam(),¶
}},¶
},¶
},¶
},¶
})¶
if err != nil {¶
log.Fatal(err)¶
}¶
fmt.Println(response)¶
}¶
```¶
¶
```java Java hidelines={1..3,5..19,-2..}¶
import com.anthropic.client.AnthropicClient;¶
import com.anthropic.client.okhttp.AnthropicOkHttpClient;¶
import com.anthropic.core.JsonValue;¶
import com.anthropic.models.messages.CacheControlEphemeral;¶
import com.anthropic.models.messages.ContentBlockParam;¶
import com.anthropic.models.messages.Message;¶
import com.anthropic.models.messages.MessageCreateParams;¶
import com.anthropic.models.messages.Model;¶
import com.anthropic.models.messages.TextBlockParam;¶
import com.anthropic.models.messages.Tool;¶
import com.anthropic.models.messages.Tool.InputSchema;¶
import com.anthropic.models.messages.ToolResultBlockParam;¶
import com.anthropic.models.messages.ToolUseBlockParam;¶
import java.util.List;¶
import java.util.Map;¶
¶
public class MultipleCacheBreakpointsExample {¶
¶
public static void main(String[] args) {¶
AnthropicClient client = AnthropicOkHttpClient.fromEnv();¶
¶
// Search tool schema¶
InputSchema searchSchema = InputSchema.builder()¶
.properties(¶
JsonValue.from(¶
Map.of("query", Map.of("type", "string", "description", "Search query"))¶
)¶
)¶
.putAdditionalProperty("required", JsonValue.from(List.of("query")))¶
.build();¶
¶
// Get document tool schema¶
InputSchema getDocSchema = InputSchema.builder()¶
.properties(¶
JsonValue.from(¶
Map.of("doc_id", Map.of("type", "string", "description", "Document ID"))¶
)¶
)¶
.putAdditionalProperty("required", JsonValue.from(List.of("doc_id")))¶
.build();¶
¶
MessageCreateParams params = MessageCreateParams.builder()¶
.model(Model.CLAUDE_OPUS_4_6)¶
.maxTokens(1024)¶
// Tools with cache control on the last one¶
.addTool(¶
Tool.builder()¶
.name("search_documents")¶
.description("Search through the knowledge base")¶
.inputSchema(searchSchema)¶
.build()¶
)¶
.addTool(¶
Tool.builder()¶
.name("get_document")¶
.description("Retrieve a specific document by ID")¶
.inputSchema(getDocSchema)¶
.cacheControl(CacheControlEphemeral.builder().build())¶
.build()¶
)¶
// System prompts with cache control on instructions and context separately¶
.systemOfTextBlockParams(¶
List.of(¶
TextBlockParam.builder()¶
.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"¶
)¶
.cacheControl(CacheControlEphemeral.builder().build())¶
.build(),¶
TextBlockParam.builder()¶
.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...]"¶
)¶
.cacheControl(CacheControlEphemeral.builder().build())¶
.build()¶
)¶
)¶
// Conversation history¶
.addUserMessage("Can you search for information about Mars rovers?")¶
.addAssistantMessageOfBlockParams(¶
List.of(¶
ContentBlockParam.ofToolUse(¶
ToolUseBlockParam.builder()¶
.id("tool_1")¶
.name("search_documents")¶
.input(JsonValue.from(Map.of("query", "Mars rovers")))¶
.build()¶
)¶
)¶
)¶
.addUserMessageOfBlockParams(¶
List.of(¶
ContentBlockParam.ofToolResult(¶
ToolResultBlockParam.builder()¶
.toolUseId("tool_1")¶
.content(¶
"Found 3 relevant documents: Document 3 (Mars Exploration), Document 7 (Rover Technology), Document 9 (Mission History)"¶
)¶
.build()¶
)¶
)¶
)¶
.addAssistantMessageOfBlockParams(¶
List.of(¶
ContentBlockParam.ofText(¶
TextBlockParam.builder()¶
.text(¶
"I found 3 relevant documents about Mars rovers. Let me get more details from the Mars Exploration document."¶
)¶
.build()¶
)¶
)¶
)¶
.addUserMessageOfBlockParams(¶
List.of(¶
ContentBlockParam.ofText(¶
TextBlockParam.builder()¶
.text("Yes, please tell me about the Perseverance rover specifically.")¶
.cacheControl(CacheControlEphemeral.builder().build())¶
.build()¶
)¶
)¶
)¶
.build();¶
¶
Message message = client.messages().create(params);¶
System.out.println(message);¶
}¶
}¶
```¶
¶
```php PHP hidelines={1..4}¶
<?php¶
¶
use Anthropic\Client;¶
¶
$client = new Client(apiKey: getenv("ANTHROPIC_API_KEY"));¶
¶
$message = $client->messages->create(¶
maxTokens: 1024,¶
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']¶
]¶
]¶
]¶
],¶
model: 'claude-opus-4-6',¶
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']¶
]¶
],¶
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']¶
]¶
],¶
);¶
¶
echo $message;¶
```¶
¶
```ruby Ruby nocheck hidelines={1..2}¶
require "anthropic"¶
¶
client = Anthropic::Client.new¶
¶
message = client.messages.create(¶
model: "claude-opus-4-6",¶
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" }¶
}¶
]¶
}¶
]¶
)¶
puts message¶
```¶
</CodeGroup>¶
¶
This comprehensive example demonstrates how to use all 4 available cache breakpoints to optimize different parts of your prompt:¶
¶
1. **Tools cache** (cache breakpoint 1): The `cache_control` parameter on the last tool definition caches all tool definitions.¶
¶
2. **Reusable instructions cache** (cache breakpoint 2): The static instructions in the system prompt are cached separately. These instructions rarely change between requests.¶
¶
3. **RAG context cache** (cache breakpoint 3): The knowledge base documents are cached independently, allowing you to update the RAG documents without invalidating the tools or instructions cache.¶
¶
4. **Conversation history cache** (cache breakpoint 4): The assistant's response is marked with `cache_control` to enable incremental caching of the conversation as it progresses.¶
¶
This approach provides maximum flexibility:¶
- If you only update the final user message, all four cache segments are reused¶
- If you update the RAG documents but keep the same tools and instructions, the first two cache segments are reused¶
- If you change the conversation but keep the same tools, instructions, and documents, the first three segments are reused¶
- Each cache breakpoint can be invalidated independently based on what changes in your application¶
¶
For the first request:¶
- `input_tokens`: Tokens in the final user message¶
- `cache_creation_input_tokens`: Tokens in all cached segments (tools + instructions + RAG documents + conversation history)¶
- `cache_read_input_tokens`: 0 (no cache hits)¶
¶
For subsequent requests with only a new user message:¶
- `input_tokens`: Tokens in the new user message only¶
- `cache_creation_input_tokens`: Any new tokens added to conversation history¶
- `cache_read_input_tokens`: All previously cached tokens (tools + instructions + RAG documents + previous conversation)¶
¶
This pattern is especially powerful for:¶
- RAG applications with large document contexts¶
- Agent systems that use multiple tools¶
- Long-running conversations that need to maintain context¶
- Applications that need to optimize different parts of the prompt independently¶
¶
</section>¶
¶
## 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 60 minutes (extended), after which they are promptly, though not immediately, deleted. Cache entries are isolated between organizations.¶
¶
For ZDR eligibility across all features, see [API and data retention](/docs/en/build-with-claude/api-and-data-retention).¶
¶
---¶
## FAQ¶
¶
<section title="Do I need multiple cache breakpoints or is one at the end sufficient?">¶
¶
**In most cases, a single cache breakpoint at the end of your static content is sufficient.** Cache writes happen only at the block you mark. Place it on the last block that stays identical across requests, and every subsequent request reads that same entry. If a later block varies per request (a timestamp, the incoming message), keep the breakpoint before it, on the last stable block.¶
¶
You only need multiple breakpoints if:¶
- A growing conversation pushes your breakpoint 20 or more blocks past the last cache write, putting the prior entry outside the lookback window¶
- You want to cache sections that update at different frequencies independently¶
- You need explicit control over what gets cached for cost optimization¶
¶
Example: If you have system instructions (rarely change) and RAG context (changes daily), you might use two breakpoints to cache them separately.¶
¶
</section>¶
¶
<section title="Do cache breakpoints add extra cost?">¶
¶
No, cache breakpoints themselves are free. You only pay for:¶
- Writing content to cache (25% more than base input tokens for 5-minute TTL)¶
- Reading from cache (10% of base input token price)¶
- Regular input tokens for uncached content¶
¶
The number of breakpoints doesn't affect pricing - only the amount of content cached and read matters.¶
¶
</section>¶
¶
<section title="How do I calculate total input tokens from the usage fields?">¶
¶
The usage response includes three separate input token fields that together represent your total input:¶
¶
```text¶
total_input_tokens = cache_read_input_tokens + cache_creation_input_tokens + input_tokens¶
```¶
¶
- `cache_read_input_tokens`: Tokens retrieved from cache (everything before cache breakpoints that was cached)¶
- `cache_creation_input_tokens`: New tokens being written to cache (at cache breakpoints)¶
- `input_tokens`: Tokens **after the last cache breakpoint** that aren't cached¶
¶
**Important:** `input_tokens` does NOT represent all input tokens - only the portion after your last cache breakpoint. If you have cached content, `input_tokens` will typically be much smaller than your total input.¶
¶
**Example:** With a 200k token document cached and a 50 token user question:¶
- `cache_read_input_tokens`: 200,000¶
- `cache_creation_input_tokens`: 0¶
- `input_tokens`: 50¶
- **Total**: 200,050 tokens¶
¶
This breakdown is critical for understanding both your costs and rate limit usage. See [Tracking cache performance](#tracking-cache-performance) for more details.¶
¶
</section>¶
¶
<section title="What is the cache lifetime?">¶
¶
The cache's default minimum lifetime (TTL) is 5 minutes. This lifetime is refreshed each time the cached content is used.¶
¶
If you find that 5 minutes is too short, Anthropic also offers a [1-hour cache TTL](#1-hour-cache-duration).¶
¶
</section>¶
¶
<section title="How many cache breakpoints can I use?">¶
¶
You can define up to 4 cache breakpoints (using `cache_control` parameters) in your prompt.¶
¶
</section>¶
¶
<section title="Is prompt caching available for all models?">¶
¶
No, prompt caching is currently only available for Claude Opus 4.6, Claude Opus 4.5, Claude Sonnet 4.6, Claude Sonnet 4.5, Claude Opus 4.1, Claude Opus 4, Claude Sonnet 4, Claude Sonnet 3.7 ([deprecated](/docs/en/about-claude/model-deprecations)), Claude Haiku 4.5, Claude Haiku 3.5 ([deprecated](/docs/en/about-claude/model-deprecations)), and Claude Haiku 3.¶
¶
</section>¶
¶
<section title="How does prompt caching work with extended thinking?">¶
¶
Cached system prompts and tools will be reused when thinking parameters change. However, thinking changes (enabling/disabling or budget changes) will invalidate previously cached prompt prefixes with messages content.¶
¶
For more details on cache invalidation, see [What invalidates the cache](#what-invalidates-the-cache).¶
¶
For more on extended thinking, including its interaction with tool use and prompt caching, see the [extended thinking documentation](/docs/en/build-with-claude/extended-thinking#extended-thinking-and-prompt-caching).¶
¶
</section>¶
¶
<section title="How do I enable prompt caching?">¶
¶
The easiest way is to add `"cache_control": {"type": "ephemeral"}` at the top level of your request body ([automatic caching](#automatic-caching)). Alternatively, include at least one `cache_control` breakpoint on individual content blocks ([explicit cache breakpoints](#explicit-cache-breakpoints)).¶
¶
</section>¶
¶
<section title="Can I use prompt caching with other API features?">¶
¶
Yes, prompt caching can be used alongside other API features like tool use and vision capabilities. However, changing whether there are images in a prompt or modifying tool use settings will break the cache.¶
¶
For more details on cache invalidation, see [What invalidates the cache](#what-invalidates-the-cache).¶
¶
</section>¶
¶
<section title="How does prompt caching affect pricing?">¶
¶
Prompt caching introduces a new pricing structure where cache writes cost 25% more than base input tokens, while cache hits cost only 10% of the base input token price.¶
¶
</section>¶
¶
<section title="Can I manually clear the cache?">¶
¶
Currently, there's no way to manually clear the cache. Cached prefixes automatically expire after a minimum of 5 minutes of inactivity.¶
¶
</section>¶
¶
<section title="How can I track the effectiveness of my caching strategy?">¶
¶
You can monitor cache performance using the `cache_creation_input_tokens` and `cache_read_input_tokens` fields in the API response.¶
¶
</section>¶
¶
<section title="What can break the cache?">¶
¶
See [What invalidates the cache](#what-invalidates-the-cache) for more details on cache invalidation, including a list of changes that require creating a new cache entry.¶
¶
</section>¶
¶
<section title="How does prompt caching handle privacy and data separation?">¶
¶
Prompt caching is designed with strong privacy and data separation measures:¶
¶
1. Cache keys are generated using a cryptographic hash of the prompts up to the cache control point. This means only requests with identical prompts can access a specific cache.¶
¶
2. Caches are organization-specific. Users within the same organization can access the same cache if they use identical prompts, but caches are not shared across different organizations, even for identical prompts.¶
¶
3. The caching mechanism is designed to maintain the integrity and privacy of each unique conversation or context.¶
¶
4. It's safe to use `cache_control` anywhere in your prompts. For caching to produce reads, place the breakpoint at the end of a stable prefix: placing it on a block that changes every request (such as a timestamp or the user's arbitrary input) writes a fresh entry each time and never hits.¶
¶
These measures ensure that prompt caching maintains data privacy and security while offering performance benefits.¶
¶
Note: Starting February 5, 2026, caches will be isolated per workspace instead of per organization. This change applies to the Claude API and Azure AI Foundry (preview). See [Cache storage and sharing](#cache-storage-and-sharing) for details.¶
¶
¶
</section>¶
<section title="Can I use prompt caching with the Batches API?">¶
¶
Yes, it is possible to use prompt caching with your [Batches API](/docs/en/build-with-claude/batch-processing) requests. However, because asynchronous batch requests can be processed concurrently and in any order, cache hits are provided on a best-effort basis.¶
¶
The [1-hour cache](#1-hour-cache-duration) can help improve your cache hits. The most cost effective way of using it is the following:¶
- Gather a set of message requests that have a shared prefix.¶
- Send a batch request with just a single request that has this shared prefix and a 1-hour cache block. This will get written to the 1-hour cache.¶
- As soon as this is complete, submit the rest of the requests. You will have to monitor the job to know when it completes.¶
¶
This is typically better than using the 5-minute cache simply because it's common for batch requests to take between 5 minutes and 1 hour to complete. Anthropic is considering ways to improve these cache hit rates and making this process more straightforward.¶
¶
</section>¶
<section title="Why am I seeing the error `AttributeError: 'Beta' object has no attribute 'prompt_caching'` in Python?">¶
¶
This error typically appears when you have upgraded your SDK or you are using outdated code examples. Prompt caching is now generally available, so you no longer need the beta prefix. Instead of:¶
<CodeGroup>¶
¶
```python Python nocheck¶
client.beta.prompt_caching.messages.create(**params)¶
```¶
¶
¶
```typescript TypeScript nocheck hidelines={1..2}¶
import Anthropic from "@anthropic-ai/sdk";¶
¶
const client = new Anthropic();¶
¶
const response = await client.beta.promptCaching.messages.create({¶
model: "claude-opus-4-6",¶
max_tokens: 1024,¶
system: [¶
{¶
type: "text",¶
text: "You are an expert on this large document...",¶
cache_control: { type: "ephemeral" }¶
}¶
],¶
messages: [{ role: "user", content: "Summarize the key points" }]¶
});¶
¶
console.log(response);¶
```¶
¶
¶
```php PHP hidelines={1..4} nocheck¶
<?php¶
¶
use Anthropic\Client;¶
¶
$client = new Client(apiKey: getenv("ANTHROPIC_API_KEY"));¶
¶
$message = $client->beta->promptCaching->messages->create(¶
maxTokens: 1024,¶
messages: [¶
['role' => 'user', 'content' => 'Summarize the key points']¶
],¶
model: 'claude-opus-4-6',¶
system: [¶
[¶
'type' => 'text',¶
'text' => 'You are an expert on this large document...',¶
'cache_control' => ['type' => 'ephemeral']¶
]¶
],¶
);¶
¶
echo $message->content[0]->text;¶
```¶
</CodeGroup>¶
Simply use:¶
<CodeGroup>¶
¶
```python Python nocheck¶
client.messages.create(**params)¶
```¶
¶
```typescript TypeScript hidelines={1..2}¶
import Anthropic from "@anthropic-ai/sdk";¶
¶
const client = new Anthropic();¶
¶
const response = await client.messages.create({¶
model: "claude-opus-4-6",¶
max_tokens: 1024,¶
system: [¶
{¶
type: "text",¶
text: "You are an expert on this large document...",¶
cache_control: { type: "ephemeral" }¶
}¶
],¶
messages: [{ role: "user", content: "Summarize the key points" }]¶
});¶
¶
console.log(response);¶
```¶
¶
```php PHP hidelines={1..4}¶
<?php¶
¶
use Anthropic\Client;¶
¶
$client = new Client(apiKey: getenv("ANTHROPIC_API_KEY"));¶
¶
$message = $client->messages->create(¶
maxTokens: 1024,¶
messages: [¶
['role' => 'user', 'content' => 'Summarize the key points']¶
],¶
model: 'claude-opus-4-6',¶
system: [¶
[¶
'type' => 'text',¶
'text' => 'You are an expert on this large document...',¶
'cache_control' => ['type' => 'ephemeral']¶
]¶
],¶
);¶
¶
echo $message->content[0]->text;¶
```¶
¶
```ruby Ruby hidelines={1..2}¶
require "anthropic"¶
¶
client = Anthropic::Client.new¶
¶
message = client.messages.create(¶
model: "claude-opus-4-6",¶
max_tokens: 1024,¶
system: [¶
{¶
type: "text",¶
text: "You are an expert on this large document...",¶
cache_control: { type: "ephemeral" }¶
}¶
],¶
messages: [¶
{ role: "user", content: "Summarize the key points" }¶
]¶
)¶
puts message.content.first.text¶
```¶
</CodeGroup>¶
¶
</section>¶
<section title="Why am I seeing 'TypeError: Cannot read properties of undefined (reading 'messages')'?">¶
¶
This error typically appears when you have upgraded your SDK or you are using outdated code examples. Prompt caching is now generally available, so you no longer need the beta prefix. Instead of:¶
¶
```typescript TypeScript¶
client.beta.promptCaching.messages.create(/* ... */);¶
```¶
¶
Simply use:¶
¶
```typescript¶
client.messages.create(/* ... */);¶
```¶
¶
</section>