Anthropic Postmortem: The Cause of Claude Code's Limit Drain and the Official Fix
Anthropic released an official postmortem on April 23, explaining the recent degradation in Claude Code's quality and the rapid drain of account Limits. The core issue stemmed from a caching bug that repeatedly dropped reasoning history, causing the AI to lose context and generate continuous cache misses. An official fix has been deployed, and the company has completely reset usage limits for all subscribers as compensation.
Official Tool Link:
The Caching Optimization Bug
In an attempt to improve efficiency, an update was deployed to clear old "thinking" (reasoning history) only for idle sessions. However, a bug caused the context to be cleared on every single turn. This caused the tool to constantly lose its memory, repeat itself, and make erratic coding decisions.
The Root Cause of the Limit Drain
Because of this continuous memory wiping, every subsequent request failed to utilize Prompt Caching, resulting in widespread Cache Misses. This flaw forced the system to transmit and consume entirely new tokens every time, which was the actual driver behind the rapid and illogical depletion of users' API limits.
Official Fix and Free Compensation
The technical issue has been completely resolved by Anthropic. As an official apology, the company announced on April 23 a complete reset of usage limits for all subscribers. Once you run the claude update command in your terminal, the tool will return to its highly efficient and cost-effective state.
Prompt
# Make sure you are on the latest version to apply the official fix (v2.1.116 or higher): claude update # Note: After updating, your API token consumption will stabilize. Anthropic has officially reset limits for all subscribers as of April 23.