Trending Now - Deepseek Unveils Major Update to AI Model Amidst High Expectations

WHO:

CNN - DeepSeek, a Chinese artificial intelligence startup based in Hangzhou.

CNN - Ivan Su, senior equity analyst at financial services firm MorningStar.

CNN - Wei Sun, principal analyst at market analysis firm Counterpoint Research.

CNN - Michael Kratsios, White House director of the office of science and technology policy.

WHAT:

CNN - DeepSeek unveiled a preview version of its new AI model, V4, on April 22nd, 2026.

CNN - The V4 model features major upgrades in reasoning and agentic abilities, capable of acting autonomously, such as writing code.

CNN - V4 enhances efficiency in processing larger numbers of tokens, the fundamental units of information for AI models.

CNN - DeepSeek's previous model, R1, was released in 2025 and delivered near industry-leading performance at a lower cost.

CNN - DeepSeek claims V4 has the best agentic coding capability among open-source models and achieves "world class" reasoning capabilities.

CNN - DeepSeek's V4 outperformed other open models in broad world knowledge but still trails behind industry leaders like Gemini.

WHERE:

CNN - DeepSeek is based in Hangzhou, China.

CNN - The V4 model is supported by Huawei's "Supernode" technology, utilizing its "Ascend 950" chips.

WHEN:

CNN - The V4 model was previewed on April 22nd, 2026.

CNN - DeepSeek's R1 model was released in 2025.

WHY:

CNN - The launch of V4 aims to compete with models from OpenAI, Anthropic, and Google.

CNN - The open-source strategy is part of China's approach to rapidly scale up AI adoption and applications across various sectors.

HOW:

CNN - DeepSeek partnered with Huawei to fulfill V4's computing needs using domestic chips from Huawei and Cambricon.

CNN - The V4 model is open source, making it available for public use, unlike most American models.

(UPDATED April 24, 2026 09:28 UTC)

DeepSeek, a Chinese AI startup based in Hangzhou, recently unveiled a preview version of its new AI model, V4, on Wednesday, showcasing significant upgrades in reasoning and autonomous capabilities, including the ability to write code. Supported by Huawei's "Supernode" technology and utilizing its "Ascend 950" chips, V4 aims to compete with leading models from OpenAI, Anthropic, and Google, while also being open source for public use. The new model is designed to enhance efficiency in processing information and claims to have the best agentic coding capability among open-source models, although it still trails behind industry leaders like Gemini.