deepseek-v4.1POST

15 项基准中仅 6 项有 V4.1 Flash 1017 成绩，领先 M3.1 Flash 4 项

用户指出 MiniMax 对比表 15 项基准只有 6 项给出 DeepSeek V4.1 Flash 1017 分数，其中 DeepSWE、Terminal Bench、NL2Repo、BabyVision 四项领先 M3.1 Flash，质疑其余未跑。

Yusuf Sadek

@yusufksadek9 天前收录原推发布于 2026.10.23

Yusuf Sadek @yusufksadek原推发布于 2026.10.23

在 X 上打开原推

@MiniMax_AI Only 6 of the 15 benchmarks have a DeepSeek V4.1 Flash-1017 score, and it's ahead of M3.1 Flash on 4 of those 6 (DeepSWE, Terminal Bench, NL2Repo, BabyVision). Is the 1017 checkpoint just too new to have run the rest

x.com/yusufksadek/status/2113457610854686972采集于 2026.10.23 12:06