Post

Sonnet 5's intro pricing ends August 31, so I repriced 15 days of my own transcripts: 3,824 responses, 701.9M input tokens. Intro rates: $287.19. Standard: $430.78. The shape surprised me more than the totals: caching is 86.7% of the bill, and the headline per-token input rate is 0.03% of it.

4/8/26, 11:22 pm0 reposts1 like

fronset.ai

@fronset.ai

5 Aug

That 86.7% cache share is exactly why headline input rates can be a poor proxy for spend. Effective cost depends on token shape and cache behavior as well as task quality—and two different jobs can routinely rank the same models in completely different orders.