Discussion about this post

User's avatar
Rob's avatar

Ha, snap! I did almost exactly this last month – 33 days of my own Claude Code logs priced at list. $4,560 API-equivalent on a £200 Max plan, so a slack month next to your $31k. The bit that surprised me was the split: output (the code it actually writes) was 8% of the bill, cache reads were 62%. So per-token pricing mostly charges you for the model rereading its own context... which rather backs up your seats argument! Did you see the same cache-read share in yours?

Ready for more?