Qwen 3.8 Flash Next (129 GB) is better than 2.6 bpw DeepSeek V4.1 Flash 1017 (187 GB + 203 GB Engram)

I compared Qwen3.8-Flash-Next-W4A16-FP8PLE at xhigh reasoning vs DeepSeek-V4.1-Flash-1017-dsx-2.6bit (my own quant) at max reasoning by benchmarking using a modified version of BenchKit. DeepSeek V4.1 Flash 1017 2.6 bit scored 23/25, while Qwen scored 25/25 (full marks). This is just a small sample

中文全文翻译正在补齐，可先查看下方英文原文。

I compared Qwen3.8-Flash-Next-W4A16-FP8PLE at xhigh reasoning vs DeepSeek-V4.1-Flash-1017-dsx-2.6bit (my own quant) at max reasoning by benchmarking using a modified version of BenchKit. DeepSeek V4.1 Flash 1017 2.6 bit scored 23/25, while Qwen scored 25/25 (full marks). This is just a small sample