WHAT NEW IN X11.02 MON

X 上的 AI 合集档案 · 22 个合集 · 64,204 条

deepseek-v4.1POST

V4.1 Flash 1017 在网络安全基准上绝对性能领先,成本约为 Astra 的 1/50

用户指出 1017 版在该安全基准上绝对性能位居前列,价格约为 Astra 的五十分之一,并认为 engram 等架构创新让小模型越级发挥。

steve hsu
@hsu_steve9 天前收录原推发布于 2026.10.23
steve hsu @hsu_steve原推发布于 2026.10.23
在 X 上打开原推

DS V4.1 Flash 1017 is among the best models in absolute performance on this cybersecurity benchmark, and is ~50x cheaper than Astra. As I have emphasized before, despite its small size, the special architecture of the model (eg innovations like engram) may allow it to punch above its weight in absolute capability as well as cost. DS are very confident that the can continue to improve with RL. The jump from the first release to 1017 is very impressive.

这条是引用着另一条写的:「DeepSeek V4.1 Flash 网络安全基准仅次于 GPT-6 Astra」

x.com/hsu_steve/status/2113440887489568894采集于 2026.10.23 12:06