老外是这样评价DeepSeek V4的景晟AI随笔

震了、慌了、嘴硬了——但订单下去了。

83.7%全球第一，老外蚌埠住了

Twitter上直接炸了：

“83.7% on SWE-Bench Verified. That would make it the best coding model in the world.”— @BridgeMind

“Stop Everything — DeepSeek V4 Might Be the Smartest Coding AI of 2026”— GoPenAI

“This is insane”— @DavidOndrej

一个编程测试打到83.7%，超越Claude Opus 4.5和GPT-5.2。老外之前没见过的阵仗。

但嘴还是硬的

Reddit上总有人要质疑一下：

“take it with a huge grain of salt, don't think these are real”

“This benchmark is probably fake. The Claude Opus 4.5 was excluded from the comparison. Why wouldn't they include it?”

意思就是：肯定假的，我们家的模型没那么容易被超。

真正怕的是这个

Tech媒体的标题才说实话：

“DeepSeek V3 wiped $1T+ from tech stocks. V4 could be even more disruptive.”

“Export controls aimed at slowing China's AI progress have backfired spectacularly.”

“China was NVIDIA's second-largest market. V4 could make that market permanently inaccessible.”

怕的不是技术强不强。

芯片不用买了

中国自己搞定了

性能对比，三句话

一个从业者的评测：

“DeepSeek is roughly 6-10x cheaper per token than comparable US frontier models.”— Abhishek Gautam

便宜6到10倍。

嘴上是质疑，身体是诚实的：

“Alibaba, ByteDance, Tencent ordered hundreds of thousands of Huawei Ascend chips”

几十万颗昇腾芯片砸下去了。

价格都涨了20%。

老外不服气，但很诚实。

这波不只是中国AI强了。

是全球AI产业链，要变天了。