310 points on Hacker News for a blog post with no product in it. Anthropic finally spelled out how Claude Code actually bills you, and the headline number is uncomfortable: everything in a session gets re-sent on every later turn. That file you read twenty turns ago is still being paid for.
The mechanics they published
Output tokens run roughly 5x the price of input. Cache reads cost 0.1x, cache writes 2x. The prompt cache expires after 1 hour on subscriptions, 5 minutes on API keys. Switching /model or effort mid-conversation busts the cache and re-prefills the whole thing at full price.
What to actually do
It’s a guide to built-in commands, not a new tool. /clear between tasks. Pick model and effort at the start and leave them. @-mention files instead of naming them, which skips a Read call. Quiet flags on noisy commands, or hand them to a subagent so only the summary comes back. /context on a fresh session to find MCP servers you never use. /compact before you walk away, while the cache is still warm.
You Might Also Like
- Pi Mono 29k Stars and a 200 Token System Prompt That Rivals Claude Code
- Anthropic cut 80 of Claude Codes System Prompt With Zero Eval Loss the new Rules of Context Engineering for Claude 5 Claude Doctor Doctor
- Claude Context Mode Might be the Best Thing Thats Happened to my Claude Code Sessions
- Anthropic Just Launched Code Review in Claude Code and 54 of prs now get Real Feedback
- Claude Replay Turns Your Anthropic Claude Code Sessions Into Shareable Video Like Replays
Top AI Product 
Leave a comment