Claude Code user reports 3.8x faster weekly usage depletion
AI summary
A Claude Code user reported that Claude Opus 5.5 consumed 3.8 times more weekly usage than the prior week, while the apparent quota remained roughly unchanged. Their trace analysis pointed to increased subagent use, duplicated context reading, and extra steps, but no independent benchmark or Anthropic confirmation was provided. The user suggested disabling subagents and starting a new session after reaching 300,000–400,000 tokens of context.
In-depth insight
The report highlights a potentially significant token-efficiency and subagent-orchestration issue for Claude Code users, along with practical mitigations to evaluate.