Anthropic acknowledged performance issues with its tool Claude Code following user complaints, identifying three product-level changes that affected user experience. The company denied allegations of intentional degradation, known as “nerfing,” stating that the underlying model remained intact.
In a blog post, Anthropic highlighted that users had reported declines in Claude Code’s performance over recent weeks, prompting an internal review. The identified issues include adjustments to the default thinking level, a cache-optimization bug, and a modification for verbosity. “We take reports about degradation very seriously. We never intentionally degrade our models,” the company stated.
As of April 20, Anthropic announced these issues have been resolved and outlined steps to prevent future occurrences. These steps include increasing the usage of the public build by staff, enhancing the code review tool, and tightening controls on system prompt changes.
Complaints from users included descriptions of Claude Code as “lazy” and “ignorant,” with Stella Laurenzo, a GitHub user and senior director at AMD, claiming the tool regressed to a point where it could not be trusted for complex engineering tasks. Many users felt validated by Anthropic’s findings, noting performance changes that coincided with the company’s updates.
Moreover, some users expressed skepticism about a recent experiment involving the removal of Claude Code from the Pro plan, which affected around 2% of new subscribers. Anthropic confirmed that the changes implemented were in response to direct user feedback and aimed at improving the overall functionality of Claude Code.
The adjustments and improvements are part of a broader response to maintain user trust and satisfaction, especially following the company’s notable $1 trillion valuation on secondary markets.
