BEIJING, Jan 29 (Reuters) - Chinese tech company Alibaba released a new version of its AI technology on Wednesday, boasting that it outperformed the acclaimed DeepSeek-V3.
The timing of the release of Qwen 2.5-Max on the first day of the Lunar New Year, a day off for many Chinese, highlights the competitive pressure the Chinese AI startup has exerted not only on international competitors but also on domestic ones.
According to Alibaba's cloud unit on its WeChat account, "Qwen 2.5-Max outperforms GPT-4o, DeepSeek-V3, and Llama-3.1-405B," which are advanced AI models from OpenAI and Meta.
The recent releases from DeepSeek, including the AI assistant powered by the DeepSeek-V3 model and the R1 model, have startled Silicon Valley, causing tech shares to drop and prompting questions about the spending plans of leading US AI firms.
The success of DeepSeek has spurred competition among Chinese companies to enhance their AI models. For instance, two days after DeepSeek-R1's launch, ByteDance, the owner of TikTok, claimed that its model surpassed OpenAI's o1 in the AIME benchmark test.
DeepSeek's previous V2 model, released last May, caused a stir in China for being open-source and cost-effective, influencing Alibaba's cloud unit to slash prices by up to 97% for various models.
Notably, Liang Wenfeng, DeepSeek's CEO, indicated in a rare interview with Chinese media that the startup prioritizes achieving AGI (artificial general intelligence) over engaging in price wars.
Contrasting DeepSeek's lean and innovative approach with the high costs and rigid structures of major Chinese tech companies, Liang suggested that traditional tech giants may not be well-suited for the future of the AI industry.
He emphasized that "Large foundational models require continued innovation, tech giants' capabilities have their limits."