When To Use This
Usetm cost to see what you spent and what KV caching saved you, without configuring a separate cost-tracking proxy. Spend comes from BillingTransaction.details (populated by the billing service at charge time, so it reconciles with the ledger). Cache savings is cached_tokens x input rate — what those tokens would have cost at full price, served at ~$0 by the cache.
Cache savings is always computed over the exact window. Per-model dollar spend is calendar-month granular (one serverless transaction per user per month): --month YYYY-MM is exact, while --window 24h|7d|30d and --start/--end report the overlapping monthly spend (noted in the output).
Usage
Examples
Last 30-day spend + cache savings, per model.
Last 7-day total (per-model defaults to current month).
Exact calendar-month spend + savings for May 2026.
Exact-window spend + cache savings.
Machine-friendly JSON for piping into other tools.
Details
Show per-model serverless spend and KV-cache savings. Spend comes fromBillingTransaction.details (authoritative, reconciles
with the ledger; calendar-month granularity). Cache savings is computed as
cached_tokens x input rate — the cost those tokens would have incurred at
full input price, served at ~$0 thanks to caching — with cached tokens read
from the serverless usage history for the exact window.
Every view shows the same shape: a per-model table plus the serverless
subtotal and total cache saved. --window 24h|7d|30d (rolling),
--month YYYY-MM (exact calendar month), and --start/--end (exact range)
just change the window. Cache savings is always window-exact; per-model
dollar spend is calendar-month granular (the finest the ledger records), so
every view except --month notes that.