Local agentic coding Benchmark : Qwen 3.8 27B (in many weights quants / cache quants / engine / reasoning effort) vs others.

In medium reasoning mode, it both scores higher than the 3.6 version, AND is very much more efficient (almost half requests needed, and a third less tokens generated) - at DeepSeek v4 Flash 3107 MXFP4 level The xhigh mode is advertised to be the best one for hard tasks. In this benchmark, howe

中文全文翻译正在补齐，可先查看下方英文原文。

In medium reasoning mode, it both scores higher than the 3.6 version, AND is very much more efficient (almost half requests needed, and a third less tokens generated) - at DeepSeek v4 Flash 3107 MXFP4 level The xhigh mode is advertised to be the best one for hard tasks. In this benchmark, howe