Loading
Loading
A live readout of how I actually use Claude Code — every day since my first session on Apr 30, aggregated locally from Claude Code’s own usage stats and the session transcripts. The dollar figure is what this would have cost at public API rates.
Almost everything is prompt cache — the same context re-read every turn, billed at a tenth of the input rate. Fresh input and generated output are a sliver of the volume but most of the cost.
Token volume by model (hover for estimated cost) — the architect-and-executors split.
Estimated cost per day — hover any bar for the detail.
Model responses by hour of day (local time).
† Cost is estimatedby applying Anthropic’s public per-token API pricing (including prompt-cache write and read rates) to the recorded token counts — the work itself ran on a flat Claude subscription, so this is the equivalent API spend, not a bill. Figures are aggregated locally across 65projects; no project names or conversation content are included. History is stitched from Claude Code’s own usage stats (through Jun 9) and the session transcripts after it, which count sessions slightly differently. Snapshot generated June 21, 2026.