deepseek-v4.1VIDEO

V4.1 Flash 1017 实测对比：DeepSWE 77.0 胜 Opus 5，Terminal-Bench 42

帖称 V4.1 Flash 1017 开源 MIT、1M 上下文，自报 DeepSWE 77.0 高于 Opus 5 的 74.0，但 Terminal-Bench 42.3 低于 51.8，并提到 $0.15/1M 输入的低谷价适合批量任务。

Sun | AI Tools Daily

@suntao802312 天前收录原推发布于 2026.10.19

Sun | AI Tools Daily @suntao8023原推发布于 2026.10.19

在 X 上打开原推

DeepSeek V4.1-Flash-1017 — open weights, MIT, 1M context. DeepSWE (their numbers): 77.0 vs Opus 5 74.0 Terminal-Bench: 42.3 vs 51.8 — not a clean sweep. $0.15 / 1M input off-peak changes the math. Cheap for bulk. Frontier for hard stuff. Would you switch any workflow? 👇

x.com/suntao8023/status/2112020362670350492采集于 2026.10.19 12:02