Chinese company DeepSeek launches its long-awaited new AI model

(AFP) Chinese startup DeepSeek on Friday unveiled a new artificial intelligence (AI) model with "drastically reduced" costs, more than a year after surprising the world with a cheap "chatbot" that matched its U.S. rivals in capabilities.

The global tech industry had been waiting several weeks for this announcement, a barometer of Chinese ambitions in the sector.

A "reasoning" model

In early 2025, the Hangzhou startup succeeded in changing the rules of the AI game with R1, a "reasoning" model that, it argued, rivaled at a lower price point Gemini, ChatGPT and Claude, U.S. giants that have received multibillion-dollar development funds.

The new DeepSeek-V4 unveiled Friday "boasts an ultra-extended context of one million words, enabling it to achieve both domestic and open source leadership in agent capabilities, world knowledge and reasoning performance," the company said in a statement.

A "preliminary version" is now available, it added.

Experts say the release of V4, based on an extended language model, or LLM, marks a "tipping point" in terms of hardware and cost.

"This addresses the long-standing problems of slower performance and higher prices associated with extended context lengths, marking a real turning point for the industry," Zhang Yi, founder of technology research firm iiMedia, told AFP.

"For end users, this will bring widespread and accessible benefits. For example, if ultra-long context support becomes a standard feature, it is expected that long text processing will go beyond cutting-edge research labs and into commercial applications," he said.

This new model of DeepSeek has two versions, DeepSeek-V4-Pro and DeepSeek-V4-Flash, the latter "a more efficient and economical option" because it has smaller parameters.

"Sputnik Moment"

V4-Pro has 1.6 trillion parameters, while V4-Flash has 284 billion, which refines the decision-making capability of the models.

"In the world's knowledge performance tests, DeepSeek-V4-Pro significantly outperforms other open source models and is only slightly outperformed by the top-tier closed source model Gemini-Pro-3.1 (from Google)," the release added.

The breakout of this Chinese company last year with the so-called "DeepSeek shock" triggered a massive sell-off in AI-related stocks, especially on Wall Street, and a rethinking of business strategies.

It was also described as a "Sputnik moment" for the sector, a colloquial expression that alludes to the instant when an industry or a society becomes uneasily aware of having fallen behind technologically in front of a competitor.

That DeepSeek chatbot then performed at a similar level to ChatGPT and other top-tier tools, but the company claimed that its development had required significantly less computing power.

However, its sudden popularity raised questions about data privacy and censorship, as the "chatbot" often refused to answer questions about sensitive topics, such as the 1989 Tiananmen crackdown.