Struct CompletionModel 

pub struct CompletionModel<T = Client> {
    pub model: String,
    pub default_max_tokens: Option<u64>,
    pub prompt_caching: bool,
    pub automatic_caching: bool,
    pub automatic_caching_ttl: Option<CacheTtl>,
    /* private fields */
}

Fields

model: String

default_max_tokens: Option<u64>

prompt_caching: bool

Enable automatic prompt caching (adds cache_control breakpoints to system prompt and messages)

automatic_caching: bool

Enable Anthropic’s automatic prompt caching (adds a top-level cache_control field to the request). The API automatically places the breakpoint on the last cacheable block and moves it forward as the conversation grows. No beta header is required.

automatic_caching_ttl: Option<CacheTtl>

TTL for automatic caching. None uses the API default (5 minutes). Set to Some(CacheTtl::OneHour) for a 1-hour TTL (requires the extended-cache-ttl-2025-04-11 beta header).

Implementations

impl<T> CompletionModel<T>

where T: HttpClientExt,

pub fn new(client: Client<T>, model: impl Into<String>) -> Self

pub fn with_model(client: Client<T>, model: &str) -> Self

pub fn with_prompt_caching(self) -> Self

Enable automatic prompt caching.

When enabled, cache_control breakpoints are automatically added to:

• The system prompt (marked with ephemeral cache)
• The last content block of the last message (marked with ephemeral cache)

This allows Anthropic to cache the conversation history for cost savings.

pub fn with_automatic_caching(self) -> Self

Enable Anthropic’s automatic prompt caching.

When enabled, a top-level cache_control: { "type": "ephemeral" } field is added to every request. Anthropic’s API automatically applies the cache breakpoint to the last cacheable block and moves it forward as the conversation grows — no beta header and no manual breakpoint management are required.

This is the recommended approach for multi-turn conversations. Use with_prompt_caching instead when you need fine-grained, per-block control over what is cached.

To use a one-hour TTL instead of the default five minutes, pass ttl: "1h" via additional_params or combine with an explicit block-level breakpoint that carries the extended TTL.

let model = client.completion_model(anthropic::completion::CLAUDE_SONNET_4_6)
    .with_automatic_caching();

Minimum cacheable prompt length

The combined prompt (tools + system + messages up to the automatically chosen breakpoint) must meet the model-specific minimum or caching is silently skipped by the API:

Model	Minimum tokens
claude-opus-4-6, claude-opus-4-5	4 096
claude-sonnet-4-6	2 048
claude-sonnet-4-5, claude-opus-4-1, claude-opus-4, claude-sonnet-4	1 024
claude-haiku-4-5	4 096

pub fn with_automatic_caching_1h(self) -> Self

Enable Anthropic’s automatic prompt caching with a 1-hour TTL.

Identical to with_automatic_caching but sets ttl: "1h" on the top-level cache_control field. Requires the extended-cache-ttl-2025-04-11 beta header to be sent with the client:

let client = anthropic::Client::builder()
    .api_key(std::env::var("ANTHROPIC_API_KEY").unwrap())
    .anthropic_beta("extended-cache-ttl-2025-04-11")
    .build()?;
let model = client.completion_model(anthropic::completion::CLAUDE_SONNET_4_6)
    .with_automatic_caching_1h();

Trait Implementations

impl<T: Clone> Clone for CompletionModel<T>

fn clone(&self) -> CompletionModel<T>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<T> CompletionModel for CompletionModel<T>

where T: HttpClientExt + Clone + Default + WasmCompatSend + WasmCompatSync + 'static,

type Response = CompletionResponse

The raw response type returned by the underlying completion model.

type StreamingResponse = StreamingCompletionResponse

The raw response type returned by the underlying completion model when streaming.

type Client = Client<AnthropicExt, T>

fn make(client: &Self::Client, model: impl Into<String>) -> Self

async fn completion( &self, completion_request: CompletionRequest, ) -> Result<CompletionResponse<CompletionResponse>, CompletionError>

Generates a completion response for the given completion request.

async fn stream( &self, request: CompletionRequest, ) -> Result<StreamingCompletionResponse<Self::StreamingResponse>, CompletionError>

fn completion_request( &self, prompt: impl Into<Message>, ) -> CompletionRequestBuilder<Self>

Generates a completion request builder for the given prompt.