On Wednesday, Chinese technology giant Alibaba 9988.HK unveiled a new version of its Qwen 2.5 artificial intelligence model, asserting that it has surpassed the esteemed DeepSeek-V3.

The timing of the Qwen 2.5-Max release, coinciding with the start of the Lunar New Year—a holiday when many Chinese workers spend time with family—highlights the competitive pressure that the rapid success of the AI startup DeepSeek has placed on both foreign and local rivals.

“Qwen 2.5-Max outshines  nearly all competitors such as GPT-4o, DeepSeek-V3, and Llama-3.1-405B,” stated Alibaba’s cloud division in an announcement shared on their official WeChat account, referencing OpenAI and Meta’s leading open-source AI technologies.

The launch of DeepSeek’s AI assistant, utilizing the DeepSeek-V3 model on January 10, along with the subsequent release of its R1 model on January 20, has startled Silicon Valley, leading to declines in tech stock prices, as investors scrutinize the significant expenditures of major AI companies in the U.S. in light of DeepSeek’s supposedly low development costs.

In response to DeepSeek’s achievements, domestic competitors have been racing to enhance their own AI model offerings.

Just two days after the introduction of DeepSeek-R1, TikTok parent company ByteDance rolled out an upgraded version of its main AI model, claiming it outperformed Microsoft-funded OpenAI’s o1 in AIME, a benchmark used to assess AI’s capability to comprehend and execute complex instructions.

This aligns with DeepSeek’s assertion that its R1 model competes with OpenAI’s o1 in several performance metrics.

DeepSeek versus local rivals

The earlier model of DeepSeek, known as DeepSeek-V2, sparked a price competition among AI models in China when it was launched last May.

DeepSeek-V2’s open-source nature and its remarkably low cost of just 1 yuan ($0.14) per 1 million tokens—units of data processed by the AI—prompted Alibaba’s cloud division to announce price reductions of up to 97% on various models.

This trend was soon followed by other Chinese tech firms, including Baidu 9888.HK, which introduced China’s first ChatGPT equivalent in March 2023, and Tencent 0700.HK, the country’s leading internet enterprise.

In a rare interview with the Chinese media outlet Waves in July, Liang Wenfeng, the mysterious founder of DeepSeek, mentioned that the startup “did not concern” itself with price wars, stating that their main aim is to achieve AGI (artificial general intelligence).

OpenAI describes AGI as autonomous systems that can outperform humans in the majority of economically significant tasks.

While expansive Chinese tech corporations like Alibaba employ hundreds of thousands, DeepSeek operates more like a research lab, primarily comprised of recent graduates and PhD students from renowned Chinese universities.

In his July discussion, Liang expressed his belief that the largest tech firms in China might not be ideally positioned for the future of the AI sector, suggesting that their substantial costs and hierarchical frameworks contrast with DeepSeek’s agile operation and informal management style.

“Large foundational models necessitate ongoing innovation; tech giants have their limits,” he remarked.

Reporting by Eduardo Baptista; Editing by Christian Schmollinger