tokenstat
tokenstat

Platform stats

What tokenstat has measured

Totals across every account that syncs: tokens, list-rate value, cache share, and how many models showed up. Nothing here is about any one person. No handles, no machines, no projects, no per-day detail. Just how much the platform has counted so far.

SampleUpdated
2
accounts syncing
of 3 registered
3
machines
reporting usage
coverage window
48 distinct days recorded

A young product and a small number. Published as-is, not dressed up.

TOKENS TRACKED
11.93B
48 distinct days recorded
USAGE VALUE
$10,486
at list rates
FROM CACHE
97.8%
of every token counted
MODELS SEEN
40
across 10 harnesses

Where the tokens went

Cache reads
11.43B
Cache writes
229.3M
Output
119.7M
Fresh input
144.3M
97.8%
came from cache

Cache reads dominate the token count and barely register in the money column. That gap is the single most under-appreciated thing about running AI coding harnesses, and it is why a raw token number on its own tells you almost nothing.

What it would have cost

At list rates
$10,486
With no cache at all
$56,988
Covered by a subscription
$10,211

None of these are anyone's bill. They are usage measured at published rates, whether or not a subscription already covered it. The point is the scale of the work, not what anybody was charged.

Volume by month

Jun 2026: 763.5M tokens, $824.14
Jul 2026: 10.40B tokens, $9,348
Aug 2026: 762.6M tokens, $313.93

Every month the platform has recorded anything in, scaled against its busiest. A short bar is a quiet month, not a missing one.