The author reports that Qwen-3.8-27B (Q4_K_S quantization, Q8_0 context) has become reliable enough to handle complex refactoring tasks autonomously, prompting them to stop using external APIs. While the model’s verbose thinking requires unsupervised operation, it consistently makes sound decisions, though it is not flawless. This performance parity with API services led the user to rely solely on the local model.