Anthropic says Claude Code did get worse — but shoots down speculation it 'nerfed' the model (www.businessinsider.com)

Anthropic has addressed user complaints regarding the recent decline in performance of its popular AI coding tool, Claude Code, acknowledging that it had indeed suffered from quality issues. In a blog post, the company clarified that these problems were not a deliberate attempt to "nerf" the model but were due to three specific product-level tweaks, including changes to its default thinking level and a cache-optimization bug. Anthropic emphasized that the underlying model remained unaffected and took immediate corrective actions to restore Claude Code's performance. The significance of this announcement lies in the AI/ML community's concerns over model regressions and user trust. The public response to these quality dips highlights the crucial relationship between user feedback and AI development. In response to the outcry, Anthropic has reset usage limits for subscribers, enhanced code review processes, and implemented stricter controls on system prompts, demonstrating a commitment to user satisfaction and model integrity. This incident illustrates the challenges faced by AI companies in maintaining performance consistency while scaling operations, particularly as they experiment with different service offerings.