Agent token efficiency · last 7 days (Sydney dates) · total $79.98

$0
08/10
$0
09/10
$0
10/10
$7
11/10
$17
12/10
$19
13/10
$37
14/10
main recipes lspn clinics ios
agent10-0810-0910-1010-1110-1210-1310-14totalwakes$/wakepeak ctx
median / p90
cache hitoutput share$/merged PR
main
opus 5.5
···$6.36$10.61$12.07$26.01$55.04310$0.1811k / 41k94%41%–
recipes
sonnet 5.5
···$0.07$5.67$7.09$7.60$20.43112$0.1818k / 63k96%42%$0.25 (82)
lspn
sonnet 5.5
······$1.81$1.817$0.2634k / 41k93%53%–
clinics
haiku 5.5
···$0.24$0.99$0.01$0.13$1.3821$0.0772k / 134k95%31%–
ios
sonnet 5.5
······$1.33$1.338$0.1720k / 39k94%49%$0.27 (5)
peak ctx = largest prompt in a wake (main cost driver; lower is better). cache hit = cached share of input tokens (higher is better; cached reads cost 10x less). output share = share of cost spent on generated tokens. $/merged PR = agent cost / PRs merged in SEAN-7/recipes (ios/* branches = ios, all others = recipes, incl. some I did myself). Updated hourly. Generated 14 Oct 22:07.