Mystery solved: Why you're burning through Claude tokens so fast and how to sort it.
retained text
I reckon I’ve finally clocked why everyone’s whingeing about losing track of their Claude usage and how it absolutely chews through tokens. It just caught my eye now that the 50% extra usage window is done and dusted. If you use `/rc` on Claude and Claude Code, whenever you reopen a chat for new bits, that chat ends up reading the entire context all over again. **How to keep a lid on your costs:** * **Finish up:** Wrap up your work completely before closing a chat. * **Start fresh:** Always crack on with a brand-new chat for the next task. * **Save it properly:** Chuck your memory and context into files—that’s exactly where it belongs. **ETA:** I know this is 'No shit Sherlock' territory for power users. But for the newbies: remember that Anthropic’s Prompt Caching dies after just 5 minutes. If you nip away, reopen a chat, and continue, Claude is forced to re-read your massive history from scratch at full whack. That’s what’s rinsing your tokens.
document fetched 2026-09-15 00:41 · canonical https://reddit.com/r/ClaudeCode/comments/1wg9w9t/mystery_solved_why_youre_burning_through_claude