# DeepSeek has released a preview version of its V4 large language model.

*genai · news · 2026-04-24 · CNBC*

## Key points

- DeepSeek released a preview version of its V4 large language model with open-source access.
- V4 is optimized for agent tools like Anthropic's Claude Code and OpenClaw, lowering inference costs.
- Huawei confirmed its Ascend AI processors can support V4, but training chip details remain unclear.
- V4's native compatibility with Chinese chips may accelerate AI sovereignty and reduce Nvidia reliance.
- The competitive framing now includes multiple Chinese open-source AI models, unlike during R1's debut.

**Companies:** Huawei, Alibaba, ByteDance, Nvidia, Counterpoint Research, Manycore Tech, DeepSeek, MiniMax, Knowledge Atlas Technology, Zhipu, SMIC, Hua Hong Semiconductor
**Countries:** China, United States

[Read the full story on CNBC](https://www.cnbc.com/2026/04/24/deepseek-v4-llm-preview-open-source-ai-competition-china.html)

---

Canonical: https://newsio.io/n/8fdfad57-8559-45aa-ac9f-01506dd5c6c7/deepseek-has-released-a-preview-version-of-its-v4-large-language-model-the-relea
Summarized by Newsio from CNBC. https://newsio.io/how-it-works