In brief: Anthropic fixed three technical bugs in Claude Code and related products that caused quality degradation between March and April; all issues resolved as of April 20, and users receive usage quota refunds.
Anthropic has identified and fixed three separate technical issues that over the past weeks led to quality degradation in Claude Code, the Claude Agent SDK, and Claude Cowork. The API remained unaffected. All issues were resolved by April 20 (version 2.1.116).
Throughout March and April, Anthropic’s engineers investigated multiple reports of degraded Claude responses. The analysis revealed three distinct root causes:
First, on March 4 the default thinking intensity in Claude Code was reduced from “high” to “medium” to reduce extremely long wait times that were freezing the interface. However, this proved to be the wrong trade-off. The change was reverted on April 7 after users made clear they prefer higher intelligence and would optionally choose lower intensity for simple tasks.
Second, a bug caused the system to delete Claude’s older thought processes from inactive sessions not just once, but on every subsequent turn. This made Claude appear forgetful and repetitive, and was corrected on April 10.
Third, on April 16 a system instruction to reduce tangential outputs was added, which degraded code quality. It was removed on April 20.
The staggered changes created the impression of widespread degradation, while internal testing initially could not reproduce the issues. In response, Anthropic is resetting usage quotas for all subscribers as of April 23.
Source: www.anthropic.com