Claude’s Performance Controversy Explained

For several weeks, a growing chorus of developers and AI power users claimed that Anthropic’s flagship models were losing their edge. Users across GitHub, X, and Reddit reported a phenomenon they described as “AI shrinkflation”—a perceived degradation where Claude seemed less capable of sustained reasoning, more prone to hallucinations, and increasingly wasteful with tokens.

The Emergence of Concerns

Critics pointed to a measurable shift in behavior, alleging that the model had moved from a “research-first” approach to a lazier, “edit-first” style that could no longer be trusted for complex engineering. The situation escalated when high-profile users and third-party benchmarks highlighted a significant trust gap, leading to Anthropic’s eventual acknowledgment of the issues.

Understanding the Technical Issues

In a recent technical post-mortem, Anthropic identified three separate product-layer changes responsible for the reported quality issues:

• Default Reasoning Effort: The shift from high to medium reasoning effort on March 4 aimed to address UI latency but resulted in a noticeable drop in intelligence for complex tasks.
• A Caching Logic Bug: A bug in the caching logic released on March 26 caused the model to lose its “short-term memory,” leading to repetitive and forgetful outputs.
• System Prompt Verbosity Limits: New instructions on April 16 to limit response verbosity inadvertently reduced coding quality evaluations by 3%.

Impact on Users and Future Safeguards

The quality issues extended beyond the Claude Code CLI, affecting the Claude Agent SDK and Claude Cowork. To regain user trust and prevent future regressions, Anthropic is implementing operational changes, including:

• Internal Dogfooding: Increased internal usage of public builds to ensure staff experiences align with user experiences.
• Enhanced Evaluation Suites: Broader evaluations for every system prompt change to isolate impacts.
• Tighter Controls: New tooling for easier audit of prompt changes and strict gating for model-specific changes.
• Subscriber Compensation: Reset usage limits for all subscribers to address token waste and performance friction.

Conclusion

This controversy highlights the delicate balance AI developers must maintain between performance, user experience, and operational efficiency. Anthropic’s commitment to transparency and ongoing evaluation is crucial in rebuilding trust with its user base.

At BlockNova, we understand the complexities of AI development and offer services tailored to your needs. Whether you’re looking for AI consultants, AI agent architecture, self-hosted LLM/AI agent hosting, or server hosting, we’re here to help you navigate the evolving landscape of AI technology.