Anthropic says Claude Code did get worse

April 24, 2026

TL;DR

Anthropic acknowledged that Claude Code's quality had recently deteriorated, validating user complaints.
Three specific product-level issues were identified: a change to the default thinking level, a bug in cache optimization, and a system prompt tweak intended to reduce verbosity.
These issues were fixed on April 20, and Anthropic has implemented measures to prevent similar problems, including increased staff usage of the public build and tighter controls on system prompt changes.
The company denied intentionally degrading the model and is resetting usage limits for subscribers.
Anthropic also recently experimented with removing Claude Code from its Pro plan for a small percentage of new users.

Continue reading the original article