https://efficienist.com/claude-code-may-be-burning-your-limits-with-invisible-tokens-you-cant-see-or-audit/
Article
-
Claims Claude Code sends more tokens than visible context suggests, raising cost concerns
-
Argues users can’t audit or see the full token usage behind the scenes
-
Points to system prompts and hidden context as likely culprits
Discussion
-
Methodologically criticized: bytes don’t correlate tightly with tokens, comparisons are invalid
-
Counter-data: ccaudit tool showed 98% of token cost is cached context — not waste, by design
-
Same issue reported simultaneously on Claude.ai, suggesting shared root cause
-
Community wants system prompt exposed for full auditability
Discuss on HN