reports say

xiāoxi

消息

indicate

chēng

称

DeepSeek (Chinese AI company)

DeepSeek

DeepSeek

V4 (model version)

V4

V4

about to / imminent

jíjiāng

即将

to release

fābù

发布

，

already

yǐ

已

targeting

zhēnduì

针对

domestically produced

guóchǎn

国产

chips

xīnpiàn

芯片

completed

wánchéng

完成

code

dàimǎ

代码

rewrite

chóngxiě

重写

。

New Words

News in English

DeepSeek V4 Launch Said to Be Imminent as Chinese AI Model Rewritten for Domestic Chips

DeepSeek V4 is expected to launch within weeks after multiple delays, with the model having been fully rewritten to run on domestic chips including Huawei Ascend. The model has native multimodal capabilities and the industry views it as a key step in advancing Chinese AI self-reliance.

According to reports, China's AI company DeepSeek's next-generation large language model V4 is expected to launch within weeks, following multiple previous delays. The primary reason for these delays was DeepSeek's decision to completely rewrite the underlying code of the model to ensure compatibility with Chinese domestically-produced chips including Huawei's Ascend and Cambricon. This move is viewed by the industry as an important initiative to advance technological self-reliance, helping to circumvent US chip export control risks. DeepSeek V4 reportedly performs outstandingly on multiple benchmark tests, possessing native multimodal capabilities supporting generation and comprehension of text, images, and video. In programming capabilities, V4 is expected to surpass existing competing models on mainstream code tests such as SWE-bench. Since late 2025, DeepSeek has garnered global attention through its V3 model, experiencing a seven-hour service outage in March this year. If V4 launches as scheduled, it will become another highly anticipated open-source large model update in April. Chinese official media also closely follows DeepSeek's development, viewing it as an important embodiment of indigenous innovation in the AI field.

More in Science & Technology