Artificial Intelligence

China's DeepSeek Launches V4 AI Model

Is DeepSeek V4 a genuine AI breakthrough or just a budget model dressed up in hype?

Published APR 24

Story

Image credit: Samuel Boivin/NurPhoto/Getty Images

The Facts

Chinese AI startup DeepSeek released preview versions of its V4 model on Friday, more than a year after its R1 model drew global attention for its performance and reportedly low training cost of under $6 million.
DeepSeek AI on X
CNBC
Al Jazeera
The V4 lineup includes two open-source models: V4-Pro, with 1.6 trillion parameters and 49 billion active parameters, and V4-Flash, with 284 billion total and 13 billion active parameters. Both feature a 1-million-token context window.
DeepSeek AI on X
Yicai on X
South China Morning Post
DeepSeek stated V4-Pro outperforms "the world's top closed-source models," trailing only Google's closed-source Gemini 3.1-Pro, while its performance falls "marginally short" of OpenAI's GPT-5.4 by an estimated three to six months.
DeepSeek AI
Reuters
Al Jazeera
V4-Pro is priced at roughly $1.80 per million input tokens and $3.50 per million output tokens. DeepSeek said prices would drop in the second half of 2026, once Huawei's Ascend 950 chips ship at scale.
The Straits Times
Nikkei Asia
The Wall Street Journal
Huawei confirmed that its "entire Ascend supernode product line now supports the DeepSeek V4 series models." However, DeepSeek did not disclose which chips were used to train V4.
CNBC
Nikkei Asia
Reuters
Following V4's release, shares of Chinese chip firms SMIC and Hua Hong Semiconductor grew 8.9% and 15.2% respectively, while rival AI firms MiniMax and Zhipu each fell around 8% to 9% in Hong Kong trading.
CNBC
South China Morning Post
The Wall Street Journal

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Narrative A

DeepSeek V4 is a genuine leap forward — offering a 1M-token context window, rock-bottom pricing and open-source access that puts frontier-level AI within reach of individual developers. The strategic pivot to Huawei and domestic Chinese chips signals that China's AI ecosystem no longer needs Nvidia to scale. Open-source models aren't just catching up; they're becoming the default.

Loong of the east on X

Isaac Kargar on X

invezz.com

Narrative B

DeepSeek V4 is still a "preview" for a reason — training problems are documented in the technical paper, and early users report nothing particularly stunning beyond a cheap price tag. U.S. export controls keep cutting-edge chips out of Chinese hands, and Silicon Valley continues to outspend Chinese rivals on top talent. Affordable isn't the same as transformative.

The New York Times

Matt Otaku on X

tae kim on X