Qwen3.6 quality gain from Q4 to Q6 quantization for coding agent

A Reddit user reports a large improvement in coding agent quality when using Q6 quantized Qwen3.6 vs Q4. The setup used llama.cpp server instead of Ollama, highlighting the impact of quantization level on local LLM performance.