请问DeepSeek官方是否针对新版本模型进行了写作文能力修正？

As the title suggests, all the web content creators we surveyed expressed that the DeepSeek v4 version has significantly improved its reasoning speed and code-building capabilities (agent development), but its Chinese writing capabilities have declined significantly, and even the experience is worse than that of the earlier R1 and V3.2 versions. Do you have any plans to optimize and fix this problem? At least, the writing level should be restored to the level of the DeepSeek v3.2 version, because most platforms are continuously removing the old version models for deep search, and other models are also gradually experiencing the problem of overfitting (including but not limited to some platforms' bizarre operations to audit the code content of the up and down lines).