Pro Max 5x quota exhausted in 1.5 hours despite moderate usage

Article Summary

Users on Claude Code’s Pro Max 5x plan reported quota exhaustion in as little as 1.5 hours despite moderate usage, triggering a high-traffic GitHub issue. Community telemetry analysis found prompt cache discounts work correctly, but background/idle sessions continue making API calls, auto-compact spikes send ~966k tokens per compaction, and the 1M context window amplifies per-call costs dramatically.

Discussion

• Boris from the Claude Code team confirmed prompt cache misses after sessions go stale (>1 hour idle) as a major factor; they are defaulting to 400k context and shipping UX nudges to /clear before resuming
• Multiple users reported switching to Codex or Cursor, citing quota issues as a breaking point
• A related root-cause GitHub issue (#46829) was closed as “not planned” by Anthropic, fueling distrust
• Workaround circulated: rolling back to Claude Code v2.1.34 and setting CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING=1 appeared to help
• Broader concern that AI providers are quietly walking back generous compute subsidies

Discuss on HN