Shane Melaugh (@shanemelaugh) on X

Why it matters

Shane Melaugh publicly questioned whether Opus 5.5 had been quietly degraded, a practice the AI community informally calls being nerfed. He asked Gemini 4 Argon to review his Claude Code sessions, and the model reportedly identified a distinct change in reasoning behavior on a specific date in early November, despite no update to the model's name. To test this, Melaugh ran a comparison on a real bug: Opus 5.5 high acted hastily, became confused, and never found a fix after hours of effort, while Opus 5.5 xhigh paused to reason and located a fix within minutes. After the post drew wide attention, Melaugh clarified that he is not an AI tools reviewer and the claim is not a rigorous benchmark; it originated from his own frustration that the model had shifted from feeling excellent to being unusable for his work. The episode highlights broader concerns about silent model changes and the difficulty of verifying them.