An update on recent Claude Code quality reports

10 hours ago

Anthropic addressed user reports of degraded Claude performance, identifying and resolving three separate issues in Claude Code, the Claude Agent SDK, and Claude Cowork, with fixes completed by April 20.
The first issue involved incorrectly setting the default reasoning effort from high to medium in Claude Code to reduce latency, which users reported made Claude less intelligent; this was reverted on April 7 to restore higher effort defaults.
A bug in prompt caching, introduced on March 26, caused Claude to continuously clear thinking history after session idle time, leading to forgetfulness and tool choice issues; it was fixed on April 10.
A system prompt change for Opus 4.7, aimed at reducing verbosity with word limits, unintentionally caused a 3% drop in intelligence; this was reverted on April 20 after evaluation.
Anthropic will implement changes to prevent future issues, including more internal use of public builds, improved Code Review tools, tighter controls on system prompt changes, and better communication via @ClaudeDevs.

Hasty Briefsbeta