User finds Gemma4 31B outperforms Qwen3.6 27B for coding

In a multi-agent coding workflow with GPT-5.5 as orchestrator, Gemma4 31B (q8_0) required less back-and-forth than Qwen3.6 27B, leading to better results. The author switched to Gemma4 as their primary coding agent after a month of frustration with Qwen.