Anthropic Reveals Causes of Claude Code Quality Degradation

Anthropic released a postmortem on the Claude Code quality degradation on April 23. The investigation covered Claude Code, Claude Agent SDK, and Claude Cowork, and the API was not affected.

The issue was not a single cause but the overlap of 3 changes, and all issues were resolved as of April 20 (v2.1.116). Anthropic noted that because these affected different time periods and user groups, the degradation appeared uneven overall.

• March 4: Lowered the default reasoning effort from high to medium. This change was meant to reduce latency but led to a perceived drop in intelligence, and was reverted on April 7.
• March 26: A bug appeared in a cache optimization that cleans up stale thinking from idle sessions, causing reasoning to keep being deleted on every subsequent turn. This made the model appear to forget and repeat itself more, and also consumed usage limits faster. The issue was fixed on April 10 (v2.1.101).
• April 16: A system prompt phrase added to shorten response length degraded coding quality, and was reverted on April 20. This change affected Sonnet 4.6, Opus 4.6, and Opus 4.7.

To prevent recurrence, Anthropic stated it will introduce expanded use of shared builds, improved code review tools, broader evals and soak periods for system prompt changes, and gradual rollouts. As of April 23, usage limits have also been reset for all subscribers.