An external investigation suggests that Anthropic's Claude Opus 5.5 has been significantly nerfed, struggling with basic coding tasks and making an average of 58 errors where Gemma 4B, an open-source model, succeeded. The investigation also noted a 50% increase in usage tokens and substantial wait times, with an average of three hours for an eight-minute task. These findings pertain to the baseline version of Claude Opus 5.5 without plugins or custom instructions. AI
IMPACT Potential decrease in user trust and performance expectations for advanced AI models, highlighting the importance of transparency in model behavior.
RANK_REASON The cluster reports on an external investigation's findings regarding a specific model's performance degradation, which falls under research into model capabilities. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
