In the world of artificial intelligence (AI), the competition is getting fiercer, especially in China. Alibaba, one of the country’s largest tech companies, has just launched a new version of its Qwen 2.5 AI model, claiming that it surpasses DeepSeek’s DeepSeek-V3, the latest version of a model that has made waves both domestically and internationally. This release, which came on the first day of the Lunar New Year when most people are celebrating with family, highlights the growing pressure DeepSeek’s rapid rise has placed not only on global competitors but also on Chinese tech giants.
The Launch of Qwen 2.5-Max
Alibaba’s cloud division recently released the Qwen 2.5-Max version of its AI model. In an announcement made via its official WeChat account, the company stated, “Qwen 2.5-Max outperforms … almost across the board GPT-4, DeepSeek-V3 and Llama-3.1-405B.” This claim sets the Qwen 2.5-Max apart as one of the most advanced AI models available, positioning itself as a competitor to some of the most prominent AI systems from OpenAI and Meta.
The timing of this release is unusual, coming on the first day of the Lunar New Year, a holiday when most Chinese workers are off and spending time with their families. This move signals how much pressure DeepSeek’s success has placed on the competition, especially with its remarkable rise in just a few short weeks.
DeepSeek’s Meteoric Rise
DeepSeek’s AI model, known as DeepSeek-V3, has made a significant impact since its release on January 10, 2025. This launch was followed by another milestone with the introduction of its R1 model on January 20, which caused shockwaves in Silicon Valley. Investors, who were already wary of the enormous costs associated with AI development, saw DeepSeek’s low-cost model and its impressive performance as a potential game-changer.
DeepSeek’s ability to deliver high-performance AI at a fraction of the cost of its competitors led to a decline in tech stocks, as companies like OpenAI were questioned about their large spending plans. DeepSeek’s success was not limited to overseas markets—its rise also ignited a rush among domestic competitors in China to update their own AI models.
Domestic Competitors Scramble to Keep Up With DeepSeek
Just days after the release of DeepSeek’s R1 model, ByteDance, the owner of TikTok, released an update to its AI model. The company claimed that its upgraded model outperformed OpenAI’s o1 model in a benchmark test called AIME, which measures how well AI understands and responds to complex instructions. This update from ByteDance echoed DeepSeek’s own claim that its R1 model was on par with OpenAI’s o1 in various performance benchmarks.
Additionally, DeepSeek’s predecessor, the DeepSeek-V2 model, had already triggered a price war in China. Released in May 2024, DeepSeek-V2 was open-source and sold at an incredibly low price—just 1 yuan (about $0.14) per 1 million tokens, which are units of data processed by AI models. This pricing strategy forced companies like Alibaba to slash prices by as much as 97% on their own AI models. Other major Chinese tech companies, including Baidu and Tencent, also followed suit with similar price cuts.
DeepSeek’s Approach: A Leaner, More Flexible Model
Despite the competitive pressure, DeepSeek’s founder, Liang Wenfeng, has stated that the company is not concerned with price wars. In a rare interview with Chinese media outlet Waves in July 2024, Liang emphasized that DeepSeek’s primary goal was not to compete on price but to achieve artificial general intelligence (AGI), a form of AI that surpasses human capabilities in most economically valuable tasks. “We do not care about price wars,” Liang said. “Our goal is AGI.”
Liang’s vision for DeepSeek stands in contrast to the practices of large tech companies like Alibaba, which operate with thousands of employees and are known for their hierarchical, top-down management structures. In his interview, Liang expressed belief that these large companies might not be well-suited to the future of the AI industry, given their high operating costs and rigid structures. DeepSeek, by comparison, operates more like a research lab, with a lean staff mainly consisting of young graduates and doctoral students from top Chinese universities.
“Large foundational models require continued innovation; tech giants’ capabilities have their limits,” Liang explained, pointing out the advantages of a smaller, more flexible operation.