Struct TokenUsage 

pub struct TokenUsage {
    pub input_tokens: Option<u64>,
    pub output_tokens: Option<u64>,
    pub cached_input_tokens: Option<u64>,
}

Raw token counts from a single LLM API response.

Contract: input_tokens is the total prompt size sent to the model (every token the model saw, regardless of cache state). cached_input_tokens is the subset of input_tokens that was served from the prompt cache. So cached_input_tokens <= input_tokens, and the billable uncached portion is input_tokens - cached_input_tokens.

Providers normalize to this shape:

• OpenAI/Compatible: prompt_tokens is already total, cached_tokens is already a subset — used directly.
• Anthropic: the API reports three DISJOINT buckets per https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching: total_input = cache_read_input_tokens + cache_creation_input_tokens + input_tokens, where Anthropic’s input_tokens is only the tokens after the last cache breakpoint. The adapter sums all three to produce the total here. cached_input_tokens is set to cache_read_input_tokens (the discount-billed subset).

Fields

input_tokens: Option<u64>

Total prompt size: uncached + cached input tokens.

output_tokens: Option<u64>

cached_input_tokens: Option<u64>

Subset of input_tokens that was served from the model_provider’s prompt cache (Anthropic cache_read_input_tokens, OpenAI prompt_tokens_details.cached_tokens).

Trait Implementations

impl Clone for TokenUsage

fn clone(&self) -> TokenUsage

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() -> TokenUsage

Returns the “default value” for a type.