V4.1 Flash 官方跑分：DeepSWE 77.0、Terminal-Bench 4.0 42.3
帖子列出 V4.1 Flash 1017 的官方成绩：DeepSWE 77.0、Automation-Bench 60.5、Terminal-Bench 4.0 42.3，并强调开源权重、1M 上下文、更快更省。
DEEPSEEK JUST MADE THE OPEN AI RACE A LOT MORE INTERESTING 💀 DeepSeek V4.1 Flash 1017 dropped today And this update is already turning heads in agentic coding 👀 On DeepSeek’s reported results: → 77.0 on DeepSWE → 60.5 on Automation-Bench → 42.3 on Terminal-Bench 4.0 But here’s what caught my attention… It’s not just scoring higher. DeepSeek says the new version can complete tasks faster while generating shorter outputs And it’s still: Open weight 1M context Cheap to run A model that’s fast, cheap and actually good at coding agents is a VERY dangerous combination 💀 The open-weight race is getting crazy And DeepSeek clearly isn’t done yet 👀