Qwen vs Gemma

Hi! Been doing some local LLM stuff, and I can't help but notice: despite vastly-superior benchmark scores, Qwen 3.6 35a3B feels... substantially less intelligent than Gemma 4 26a4B (QAT). In terms of prompt adherence, output coherence, and just general "sanity", Gemma seems head-and-shoulders ahead

中文全文翻译正在补齐，可先查看下方英文原文。

Hi! Been doing some local LLM stuff, and I can't help but notice: despite vastly-superior benchmark scores, Qwen 3.6 35a3B feels... substantially less intelligent than Gemma 4 26a4B (QAT). In terms of prompt adherence, output coherence, and just general "sanity", Gemma seems head-and-shoulders ahead