Struct TokenUsage 

pub struct TokenUsage {
    pub prompt_tokens: u64,
    pub completion_tokens: u64,
    pub total_tokens: u64,
    pub context_window: u64,
    pub cache_read_input_tokens: u64,
    pub cache_creation_input_tokens: u64,
    pub reported_cost_usd: Option<f64>,
    pub reasoning_tokens: Option<u64>,
}

Token usage statistics from a model response.

Fields

prompt_tokens: u64

Number of tokens in the prompt/input.

For providers with prompt caching (Anthropic), this is the non-cached prefix only — the tokens after the last cache breakpoint. The cached portion is reported separately in Self::cache_read_input_tokens / Self::cache_creation_input_tokens, so the true input total is the sum of all three. Pricing those three buckets at the same rate over- or under-counts cost; see crate::outcome::ModelProfile::usd_per_success.

completion_tokens: u64

Number of tokens in the completion/output.

total_tokens: u64

Total tokens (prompt + completion).

context_window: u64

Model’s maximum context window size.

cache_read_input_tokens: u64

Prompt-cache hit: input tokens read from a previously written cache entry. Billed at ~0.1× the base input rate. 0 when the provider has no prompt caching, caching was disabled, or nothing hit. (Anthropic usage.cache_read_input_tokens.)

cache_creation_input_tokens: u64

Prompt-cache write: input tokens written into the cache this request. Billed at ~1.25× (5-minute TTL) or ~2× (1-hour TTL) the base input rate. 0 when caching is off or nothing was written. (Anthropic usage.cache_creation_input_tokens.)

reported_cost_usd: Option<f64>

The provider’s own dollar figure for this call, when its response carried one (OpenRouter’s usage.cost, which CAR asks for). None means the provider sent no figure, never that the call was free; the call is then priced from the catalog (billing::call_cost). Omitted from the wire when absent.

reasoning_tokens: Option<u64>

Reasoning tokens the provider says it spent, a subset of Self::completion_tokens (already counted and billed there, not additional). Read from OpenAI chat usage.completion_tokens_details.reasoning_tokens and Responses usage.output_tokens_details.reasoning_tokens, on non-streaming calls. None means the provider reported no figure — Anthropic, Google, Bedrock, local models and streamed calls today — never that no reasoning happened. Omitted from the wire when absent.

Trait Implementations

impl Clone for TokenUsage

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TokenUsage

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for TokenUsage

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for TokenUsage

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl JsonSchema for TokenUsage

fn schema_name() -> String

The name of the generated JSON Schema.

fn schema_id() -> Cow<'static, str>

Returns a string that uniquely identifies the schema produced by this type.

fn json_schema(generator: &mut SchemaGenerator) -> Schema

Generates a JSON Schema for this type.

fn is_referenceable() -> bool

Whether JSON Schemas generated for this type should be re-used where possible using the $ref keyword.

impl Serialize for TokenUsage

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.