[Claude Code Experienced Temporary Downgrade in Intelligence Due to Parameter and Prompt Adjustments, Fully Rolled Back and Fixed by Official Team] According to the Anthropic blog, Anthropic recently acknowledged a decline in code quality for Claude Code, attributed to three product and prompt changes: 1. On March 4, the default reasoning effort was lowered from high to medium, resulting in reduced intelligence. This was restored to high/xhigh on April 7. 2. On March 26, a bug in the clear_thinking_20251015 logic introduced for cache optimization caused prolonged idle periods to continuously lose reasoning history, leading to forgetfulness and repetition. This was fixed in version v2.1.101 on April 10. 3. On April 16, a word count compression feature was added with a limit of '≤25/100 words.'