# DeepSeek 發表其新 AI 模型 V4 的預覽版本，該模型專為華為晶片運行而設計。

*genai, semiconductor · news · 2026-04-24 · Devdiscourse*

## Key points

- DeepSeek 的 V4 AI 模型專為華為昇騰 AI 晶片運行而設計。
- V4 針對 Claude Code 和 OpenClaw 等代理框架設計，而非傳統聊天機器人。
- 轉向華為晶片代表擺脫對美國晶片技術的依賴。
- V4 的專業版在 STEM 和競賽程式設計方面表現優異，但落後於 Gemini 3.1 Pro 和 GPT-5.4。

Chinese startup DeepSeek unveiled a preview version of its new AI model, V4, engineered to operate on Huawei chips, signaling China's strategic move towards AI self-sufficiency. DeepSeek has designed V4 to integrate with agent frameworks like Claude Code and OpenClaw, shifting focus from chatbots to models performing complex tasks. The model comes in two forms: Pro and Flash, with Pro offering superior capabilities in areas like STEM and competitive programming, though still trailing behind top closed-source models such as Google's Gemini 3.1 Pro and OpenAI's GPT-5.4 in certain respects. Critically, V4's adaptation for Huawei's Ascend AI chips marks a significant departure from previous collaborations with U.S. chipmakers, aligning with a domestic strategy. Despite technological gaps, this partnership leverages DeepSeek's market position and underscores a tangible shift towards AI infrastructure independence, albeit under U.S. export restrictions impacting computing capacity. (With inputs from agencies.)

**Companies:** DeepSeek
**Countries:** China

[Read the full story on Devdiscourse](https://www.devdiscourse.com/article/technology/3886057-deepseek-unveils-v4-ai-model-a-leap-towards-chinas-self-sufficient-ai-ecosystem)

---

Canonical: https://newsio.io/zh-TW/n/0673c58b-7612-436d-9877-01d204e7dc9b/deepseek-ai-v4-v4-ai
Summarized by Newsio from Devdiscourse. https://newsio.io/how-it-works