# DeepSeek released its newest large language model in a preview capacity.

*genai · news · 2026-04-24 · Fortune*

## Key points

- DeepSeek's V4-Pro model has 1.6 trillion parameters and 1 million token context length.
- DeepSeek trained V4 on Huawei's Ascend AI processors, moving away from reliance on Nvidia chips.
- V4-Pro costs $3.48 per million tokens, undercutting OpenAI and Anthropic by nearly tenfold.
- DeepSeek is seeking a $20 billion valuation, primarily to retain top AI talent.
- U.S. export controls accelerated Chinese innovation in chip efficiency, influencing DeepSeek's pricing edge.

**Companies:** Semiconductor Manufacturing International Corporation, DeepSeek, Huawei, Minimax, Knowledge Atlas, Moonshot AI, Alibaba
**Countries:** China, United States

[Read the full story on Fortune](https://fortune.com/2026/04/24/deepseek-v4-ai-model-price-performance-china-open-source/)

---

Canonical: https://newsio.io/n/92a83a97-ea6f-46ed-9d45-6a0d509f080e/deepseek-released-its-newest-large-language-model-in-a-preview-capacity-the-rele
Summarized by Newsio from Fortune. https://newsio.io/how-it-works