PANews reported on April 24th that, according to the Anthropic blog, Anthropic recently acknowledged a decline in Claude Code code quality due to three product and prompt word changes: First, the default reasoning effort was downgraded from high to medium on March 4th , leading to a decrease in intelligence; this was restored to high/xhigh on April 7th . Second, the clear_thinking_20251015 logic introduced on March 26th to optimize caching had a bug , causing inference history to be continuously lost during prolonged idle periods, leading to forgetting and duplication; this was fixed in v2.1.101 on April 10th . Third, the addition of a "≤25/100 words" system prompt for word compression on April 16th had an approximately 3% negative impact on code quality ; this was rolled back on April 20th . Anthropic stated that the API inference layer was unaffected and that usage quotas for all subscribers have been reset.