Full Breakdown
Anthropic Addresses Quality Concerns with Claude Code
4/24/2026, 11:27:09 AM
Recent Quality Issues Identified
Anthropic, the AI company behind Claude Code, has acknowledged a decline in the quality of its AI model following user complaints. In a blog post published on April 20, 2026, the company confirmed that three specific changes made in March and April contributed to a perceived degradation in performance. Users reported that Claude Code seemed less capable of complex reasoning and more prone to errors, leading to widespread frustration among developers and AI users.
Core Changes Leading to Degradation
The issues stemmed from three main adjustments:
1. Default Reasoning Effort Adjustment: On March 4, Anthropic changed Claude Code's default reasoning effort from high to medium to address latency issues that caused the user interface to appear frozen. This decision was reversed on April 7 after users expressed a preference for higher intelligence levels.
2. Caching Logic Bug: A change implemented on March 26 aimed to optimize caching by clearing older reasoning from sessions that had been idle for over an hour. However, a bug caused this clearing to occur on every turn, resulting in the model appearing forgetful and repetitive. This issue was fixed on April 10.
3. System Prompt Verbosity Limits: On April 16, a new system prompt was introduced to limit verbosity, which inadvertently led to a 3% drop in coding quality evaluations. This change was reverted on April 20.
These adjustments collectively created the impression that Claude Code had become less intelligent, a phenomenon described by users as "AI shrinkflation."
User Reactions and Criticism
The response from users has been mixed. Many expressed validation upon hearing Anthropic's acknowledgment of the issues. For instance, Stella Laurenzo, a senior director at AMD, shared her findings on GitHub, indicating a significant drop in Claude's reasoning depth. Other users reported that the model's performance had deteriorated to the point where it could not be trusted for complex engineering tasks.
Critics have also raised concerns about the overzealous application of the Acceptable Use Policy (AUP) in the latest version, Opus 4.7. Users reported numerous false positives where legitimate requests were flagged as violations, hampering productivity. One user, Golden G Richard III, noted that basic editing tasks were being rejected, questioning the balance between security and usability.
Official Statements and Future Safeguards
In response to the backlash, Anthropic has committed to several operational changes to prevent similar issues in the future. These include requiring a larger share of internal staff to use public builds of Claude Code, enhancing evaluation processes for system prompt changes, and implementing tighter controls on modifications.
Anthropic has also reset usage limits for all subscribers as a form of compensation for the performance issues experienced. The company aims to maintain transparency with users through a new account on social media platform X, where it will share insights into product decisions.
Conclusion
Anthropic's recent acknowledgment of quality issues with Claude Code reflects the challenges faced by AI companies in balancing performance, user experience, and safety. As the company implements corrective measures, it remains to be seen how effectively it can restore user trust and improve the functionality of its AI models.
