China Escalates AI Race wi Alibaba’s Qwen 2.5

Alibaba Group Holding Ltd. has intensified the artificial intelligence competition by unveiling benchmark scores that position its latest AI model, Qwen 2.5 Max, as a world leader. The company claims the upgraded model outperforms Meta’s Llama and DeepSeek’s V3 model in several key tests, according to an announcement by Alibaba Cloud on WeChat. Like Tencent Holdings Ltd. and Baidu Inc., Alibaba has been heavily investing in its cloud computing segment, vying to attract China’s AI developers to its platform. This latest release comes amid a heated contest in the Chinese AI landscape, where companies are pushing the limits of innovation despite restrictions on access to advanced AI chips due to US sanctions.

DeepSeek’s Rapid Rise Sparks Fierce Competition

DeepSeek, a 20-month-old startup based in Alibaba’s home city of Hangzhou, became an international sensation this week, setting a new benchmark for AI performance. Alibaba, in its announcement, appeared to measure itself directly against DeepSeek, highlighting its superiority in coding and user query benchmarks. The company also shared data suggesting that Qwen 2.5 Max outperforms OpenAI and Anthropic’s models in specific evaluations. In recent months, cloud service providers such as Alibaba and Tencent have aggressively cut prices to attract more users. The growing influence of DeepSeek—alongside several other AI startups that have secured unicorn valuations—has fuelled this price war, making AI services more accessible.

China’s AI Innovation: High Efficiency at Lower Costs

China’s approach to artificial intelligence has ushered in an era of “frugal innovation,” producing AI models that require significantly fewer computational resources than those developed in the West. This trend was further reinforced with the release of DeepSeek V3 and, more recently, Alibaba’s Qwen 2.5. Alibaba asserts that Qwen 2.5 Max delivers top-tier performance, outpacing DeepSeek V3 in widely recognised coding and user query benchmarks. The company has pledged to continue scaling the model, enhancing its pre-training, and investing in reinforcement learning to improve reasoning capabilities.

Despite restrictions on high-end AI chip imports, Chinese tech giants such as Alibaba, DeepSeek, and Tencent have devised efficient ways to optimise their models. The outcome? Open-source AI systems that rival OpenAI’s models in reasoning benchmarks while delivering nearly twice the speed—trained at less than a tenth of the computational cost.

AI Race Heats Up Amid DeepSeek’s Disruption

The timing of Qwen 2.5 Max’s launch—on the first day of the Lunar New Year, a national holiday—underscores the pressure DeepSeek’s meteoric rise has placed on both Chinese and international AI developers. The rapid rise of DeepSeek has sent shockwaves through Silicon Valley. The launch of DeepSeek V3 on 10 January, followed by the release of its R1 model on 20 January, has prompted investor concerns over the massive spending required by US-based AI firms. The startup’s ability to develop and run AI at dramatically lower costs has caused tech stocks to decline and raised questions about the sustainability of heavy investment in AI infrastructure.

Domestic AI Firms Scramble to Keep Up

DeepSeek’s unexpected success has triggered a race among Chinese AI firms to refine and upgrade their models. Just two days after the release of DeepSeek-R1, ByteDance—the parent company of TikTok—unveiled an improved version of its flagship AI model. ByteDance claims this model surpasses OpenAI’s o1 in AIME, a benchmark assessing AI’s ability to comprehend and respond to complex instructions. This move echoes DeepSeek’s own claim that its R1 model rivals OpenAI’s o1 in multiple key performance benchmarks. As China’s AI race intensifies, companies are innovating at breakneck speed, challenging the dominance of Western AI firms while driving the next generation of cost-efficient, high-performance models.