Any opinion about Qwen3.6-27B@BF16 vs Step3.7@IQ4_XS?

The discussion compares Qwen3.6-27B@BF16, a dense model operating at full precision, with Step3.7@IQ4_XS, a mixture of experts (MoE) model featuring 7x more parameters but suboptimal quantization. The user expresses frustration with Qwen-Coder-Next@Q8's performance in programming tasks, highlighting issues with suboptimal code generation and maintainability, raising questions about the decision-making capabilities of these models in practical applications. This comparison is relevant for practitioners evaluating model trade-offs between precision, memory usage, and practical output quality in AI programming tasks.