The models teams run in production, and what they pay for them.

14.6%

of tokens on deepseek-v4.1-flash

+4.8pp

jev, week over week

$0.47

blended cost per million tokens

73%

cut from the bill by caching