Struct TokenCounts 

pub struct TokenCounts {

    pub input_tokens: i64,
    pub output_tokens: i64,
    pub reasoning_tokens: i64,
    pub cache_read: i64,
    pub cache_creation: i64,
    pub cache_creation_5m: i64,
    pub cache_creation_1h: i64,
    pub web_search_requests: i64,
    pub total: i64,
    pub above_input: i64,
    pub above_output: i64,
    pub above_reasoning: i64,
    pub above_cache_read: i64,
    pub above_cache_creation_5m: i64,
    pub above_cache_creation_1h: i64,
}

Normalized token counts extracted from provider-specific usage data.

cache_creation is the total across TTL variants. For Claude Code, the cache_creation_5m / cache_creation_1h split reflects Anthropic’s two cache TTL tiers (5 minutes default, 1 hour extended — the latter is ~60% more expensive per write). Providers that don’t split TTL (Codex, Gemini, older Claude Code records) get all cache_creation tokens into the 5m bucket.

Invariant: cache_creation == cache_creation_5m + cache_creation_1h when the split data is available.

reasoning_tokens carries the model’s “thinking” budget emitted as part of the assistant turn but billed separately from user-visible output. Populated by Gemini (thoughts_tokens), Codex (reasoning_output_tokens), and Copilot (reasoning_output_tokens after session::copilot::parse_copilot_events normalises it). Claude has no equivalent and leaves this at 0.

Fields

input_tokens: i64

Non-cached prompt tokens (cached reads are excluded; see cache_read).

output_tokens: i64

User-visible completion tokens, excluding reasoning.

reasoning_tokens: i64

Model “thinking” tokens, billed separately from output_tokens.

cache_read: i64

Prompt tokens served from the provider’s prompt cache.

cache_creation: i64

Total cache-write tokens across all TTL tiers.

cache_creation_5m: i64

Cache-write tokens at the default 5-minute TTL.

cache_creation_1h: i64

Cache-write tokens at the extended 1-hour TTL.

web_search_requests: i64

Server-side web-search requests (Claude server_tool_use). Billed per query (not per token) at the model’s web-search rate, so it is tracked here but excluded from total.

total: i64

Sum of the billed buckets used for cost and display.

above_input: i64

Slice of input_tokens from requests whose own prompt context exceeded the model’s context-tier threshold (see the above_tier object written by the usage parsers). Always a subset of the field it mirrors — never additional tokens — so displays keep using the totals above while calculate_cost bills this slice at the tier rate.

above_output: i64

Above-threshold slice of output_tokens.

above_reasoning: i64

Above-threshold slice of reasoning_tokens.

above_cache_read: i64

Above-threshold slice of cache_read.

above_cache_creation_5m: i64

Above-threshold slice of cache_creation_5m.

above_cache_creation_1h: i64

Above-threshold slice of cache_creation_1h.

Implementations

impl TokenCounts

pub fn has_activity(&self) -> bool

Whether any billed bucket carries a nonzero count.

The single source of truth for “did this usage do anything” — used to decide whether a date counts as active. New buckets added above must be reflected here so no activity check silently misses them.

Trait Implementations

impl Debug for TokenCounts

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

Formats the value using the given formatter.

impl Default for TokenCounts

fn default() -> TokenCounts

Returns the “default value” for a type.