Struct Usage 

pub struct Usage {
    pub prompt: u32,
    pub completion: u32,
    pub total: u32,
    pub prompt_tokens_details: Option<PromptTokensDetails>,
    pub completion_tokens_details: Option<CompletionTokensDetails>,
}

Token usage statistics in OpenAI’s shape.

The cache and reasoning shares ride in OpenAI’s own breakdown objects, so a stock client keeps reading the three counts and a cost-aware one finds the shares where it already looks. A share the runner did not report is absent, never a zero: None means “not measured”, Some(0) “measured zero”.

Reasoning is counted the opposite way on each side of this type. An embacle TokenUsage keeps it out of completion_tokens (additive, the way its pricing bills it); OpenAI’s completion_tokens includes it and completion_tokens_details.reasoning_tokens is a breakdown of that count. The From conversion folds reasoning into completion_tokens, so a client pricing by completion_tokens bills the reasoning, and one taking completion_tokens - reasoning_tokens gets the visible output.

Fields

prompt: u32

Tokens in the prompt, cached reads and writes included

completion: u32

Tokens in the completion, reasoning tokens included

total: u32

Total tokens, never below prompt_tokens + completion_tokens

prompt_tokens_details: Option<PromptTokensDetails>

The cached share of prompt_tokens, when the runner reported one

completion_tokens_details: Option<CompletionTokensDetails>

The reasoning share of completion_tokens, when the runner reported one

Trait Implementations

impl Debug for Usage

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

Formats the value using the given formatter.

impl From<&TokenUsage> for Usage

fn from(usage: &TokenUsage) -> Self

Converts to this type from the input type.

impl Serialize for Usage

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

where __S: Serializer,

Serialize this value into the given Serde serializer.