V4.1 Flash 1017 跑自建 mini-bench，将与 GPT-6 Luna、Haiku 5.5 对比

Ibragim @ibragim_bad原推发布于 2026.10.17

在 X 上打开原推

I ran the new DeepSeek-V4.1-Flash-1017 on my-mini-bench both through @pidotdev and @DeepSeekHarness later will post how it actually behaves and differs from gpt-6-luna and haiku-5.5 my-mini-bench is just tasks from my day-to-day work: tasks are well-defined and not that hard. I will link in reply how I built them + full leaderboard. the point is to see how cost-efficient a setup is, not to rank models.