New AI model can outperform DeepSeek - What is known

What is known about the new AI model

The unusual timing of the Qwen 2.5-Max release - on the first day of the Chinese New Year, when most Chinese are on vacation and spending time with their families - indicates the growing pressure from DeepSeek's rapid rise. Over the past three weeks, the startup has caused a stir not only among foreign competitors but also among local tech giants.

"Qwen 2.5-Max outperforms GPT-4o, DeepSeek-V3, and Llama-3.1-405B in almost all aspects," Alibaba's cloud division said in an official announcement on WeChat, referring to the most advanced open source AI models OpenAI and Meta.

The January 10 release of DeepSeek, an AI assistant based on the DeepSeek-V3 model, and the January 20 release of DeepSeek-R1 shook Silicon Valley. The stock price of tech companies fell as the claimed low costs of developing and using the Chinese model made investors doubt the multibillion-dollar spending of the largest AI companies in the US.

However, the success of DeepSeek has sparked a race among Chinese competitors to update their own AI models.

Two days after the release of DeepSeek-R1, TikTok owner ByteDance presented an update to its flagship AI, claiming that it outperforms OpenAI o1 (supported by Microsoft) in the AIME test, which assesses the ability of models to understand and execute complex instructions. This confirms DeepSeek's claim that its R1 is comparable to OpenAI o1 in a number of metrics.

DeepSeek vs. local competitors

The predecessor of DeepSeek-V3, the DeepSeek-V2 model, sparked a "price war" in the AI market in China after its release in May 2023.

The fact that DeepSeek-V2 was open-source and extremely cheap (only $0.14 for 1 million tokens) led Alibaba Cloud to announce a price cut of up to 97 percent on its AI models. It was followed by other Chinese tech giants, including Baidu (9888.HK), which launched the first Chinese equivalent of ChatGPT in March 2023, and China's largest internet company Tencent (0700.HK).

DeepSeek founder Liang Wenfeng said in a rare interview with the Chinese publication Waves in July that the startup is "not interested in a price war" as its main goal is to achieve AGI (artificial general intelligence).

OpenAI defines AGI as autonomous systems that outperform humans in most economically significant tasks.

While Alibaba and other Chinese giants have hundreds of thousands of employees, DeepSeek operates more like a research laboratory. The team includes graduates and doctoral students from leading Chinese universities.

Liang argues that China's largest tech corporations may not fit into the future of the AI industry, as their high costs and hierarchical structure hinder flexibility and speed of innovation.

"Large basic models require constant development, and the capabilities of tech giants are not unlimited," he emphasized.