Cached tokens billed at the cached rate
Coding agents resend large system prompts and file context on most turns. The model providers meter cached vs. fresh input separately — Newton's passes that distinction straight through, so repeated context isn't billed as if it were new every time.