0% 5h quota used ·
resets in 2h
live · 08/09/2026 12:04 CEST
The books
Every run priced from #1. Ceilings are per-run and per-day per country, with $540 a day as the fleet backstop.
API-equivalent
$4,536
across 516 runs · what this would have cost on the API
Cash spent
$0.53
real money on prepaid providers · $0 on the subscription
Agent time
98h 36m
13,170 turns
Days over the ceiling
0
ceiling $540 a day
Runs over the ceiling
0
each against its own country's per-run ceiling
Spend per day
red where the day broke $540 — the fleet backstop, the sum of every country's daily ceilingWhat the money went on
by task, all of the selectionBy workspace
By model
script runs cost nothing by designWaiting for a slot
the cost of the concurrency cap, in time rather than money — measured on 140 agent runs since it started being recorded
Runs that waited
13
of 140 measured · 9%
Time spent queued
5h 03m
work that was ready and could not start
| Task | Runs that waited | Total | Worst |
|---|---|---|---|
| de:netzbetreiber-replies | 10 | 4h 54m | 36m |
| _global:monthly-report | 1 | 5m | 5m |
| de:wasserversorger-enrich | 1 | 2m | 2m |
| _global:pr-review | 1 | 1m | 1m |
Month by month
| Month | Runs | Failed | Turns | Agent time | Cost | Per run |
|---|---|---|---|---|---|---|
| 2026-09 | 104 | 3 | 2,944 | 28h 08m | $1,531 | $14.72 |
| 2026-08 | 262 | 12 | 8,651 | 60h 16m | $2,612 | $9.97 |
| 2026-07 | 150 | 47 | 1,575 | 10h 11m | $393 | $2.62 |
Runs that cost far more than their task usually does
over 3× the task median, and over a dollarNothing here was actually invoiced: runs go through the Claude subscription, and cost_usd is what the same tokens would have cost on the API. That is the point — 71 of these 516 runs are deterministic scripts and cost nothing at all, and the gap between those two numbers is the argument for moving work out of agent turns.