Is Claude Opus 4.7 Underperforming? Users Report Significant Decline in Intelligence and Performance

There's a growing sentiment among users that Claude Opus 4.7 has experienced a noticeable degradation in performance and intelligence. Many feel that the model is now less capable than its predecessor, Opus 4.6, leading to a more frustrating and less productive user experience.

Performance Decline and Reasoning Issues

One of the most frequently cited concerns is the perceived "nerfing" of Opus 4.7. Users report that the model exhibits shallower thinking, struggles with deep reasoning tasks, and has difficulty retaining context from previous interactions. This leads to instances where it fails to push back on incorrect inputs, readily changes its opinions, and cannot recall information discussed just moments before. Calculations made by the model are also reported to be incorrect, leading to a sense that it is hallucinating or providing unreliable information.

The model is also noted to be slower and less responsive, contributing to user frustration. There's speculation that this decline might stem from secretive tuning down of its RAG (Retrieval Augmented Generation) capabilities or changes to how it manages context. Furthermore, users observe that the model's resource usage seems to deplete more quickly.

Strategies for Mitigating Limitations

In light of these challenges, one user shared a productive approach: employing an external AI like Codex to audit Claude's actions. This involves using Codex to independently reason about a problem or review Claude's outputs, effectively acting as a quality control layer. This method has reportedly exposed significant shortcomings in Claude's reasoning, highlighting how "dumb" it can be without external validation.

The widespread agreement on these observations suggests a consistent experience across a user base that has come to expect higher performance from the Opus series.