Token and cost data

Token fields

The dashboard’s headline token metric is:

total_tokens = input_tokens + output_tokens

Cache and reasoning fields are separate counters:

Field	Meaning
input_tokens	New prompt content processed for a call.
output_tokens	Content returned by the model.
cache_read_tokens	Previously cached context reused by a later call.
cache_write_tokens	Context added to cache for later reuse.
reasoning_tokens	Tokens used for model reasoning before the visible response.

The Composition chart shows all five counters. It always uses raw token counts because the export does not provide a separate cost for each category.

Cost estimates

The dashboard estimates list-price cost from total_nano_aiu:

estimated_cost_usd = total_nano_aiu / 1e11

This is an estimate. Plan allowances, discounts, taxes, currency, and missing cost data can make it differ from an invoice.

cost_data_calls shows how many calls in an aggregated row had recorded cost data. The Cost data coverage KPI identifies incomplete coverage. Do not treat a zero or missing estimate as proof that usage was free.