Struct UsageInfo 

pub struct UsageInfo {
    pub input_tokens: u32,
    pub cache_creation_input_tokens: u32,
    pub cache_read_input_tokens: u32,
    pub output_tokens: u32,
    pub server_tool_use: ServerToolUse,
    pub service_tier: String,
    pub cache_creation: Option<CacheCreationDetails>,
    pub inference_geo: Option<String>,
    pub iterations: Vec<UsageIteration>,
    pub speed: Option<String>,
    pub output_tokens_details: Option<OutputTokensDetails>,
}

Usage information for the request.

These counters are accumulated roll-ups, not snapshots. The CLI sums usage field-by-field across every API call (“iteration”) in the turn’s tool-use loop, so on a turn with N iterations each counter is the sum of all N. This is the right number for cost, and the wrong number for context occupancy: cache_read_input_tokens in particular re-counts the cached context on every iteration and can exceed the model’s context window several times over on tool-heavy turns. To estimate context occupancy, use the last entry of iterations instead (the CLI itself does exactly this).

Note: the result frame’s usage covers the main agent only — the subagent (Task / sidechain) token rollup the CLI renders as <subagent_tokens> / <agent_count> is not carried here or anywhere else on the wire. Accumulate it from Task tool results with SubagentUsageRollup.

Fields

input_tokens: u32

Sum of fresh (uncached) input tokens across all iterations of the turn.

cache_creation_input_tokens: u32

Sum of cache-write input tokens across all iterations of the turn.

cache_read_input_tokens: u32

Sum of cache-read input tokens across all iterations of the turn. Re-counts the cached context every iteration — not a measure of context occupancy; see the type-level docs.

output_tokens: u32

Sum of output tokens across all iterations of the turn.

server_tool_use: ServerToolUse

service_tier: String

cache_creation: Option<CacheCreationDetails>

Cache creation breakdown

inference_geo: Option<String>

Inference geography (e.g., “not_available”)

iterations: Vec<UsageIteration>

Per-iteration usage breakdown within this turn — one entry per API call in the tool-use loop, in order. The last entry reflects the final API call and is what the CLI reads for context-size estimates.

speed: Option<String>

Speed tier (e.g., “standard”)

output_tokens_details: Option<OutputTokensDetails>

Output-token breakdown (CLI 2.1.232+): currently the thinking-token share of output_tokens.

Trait Implementations

impl Clone for UsageInfo

fn clone(&self) -> UsageInfo

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for UsageInfo

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for UsageInfo

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for UsageInfo

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

where __S: Serializer,

Serialize this value into the given Serde serializer.