The Qwen 3.8:27b-BF16 model represents a regression for non-coding tasks like document analysis and complex reasoning, according to user feedback. While the previous Qwen 3.6:27b-BF16 was used daily for such purposes, the newer version exhibits excessive thinking behavior, leading to inefficient web searches and unfocused exploration of irrelevant details. Users report that even with thinking disabled and web tools enabled, the model performs numerous web search iterations without productive outcomes. This feedback is echoed across self-hosted LLM communities, where other users have expressed similar concerns about the model's performance.