How Alibaba’s Qwen 2.5‑Max Challenges GPT‑4o and Redefines China’s AI Race

President Xi Jinping recently urged private tech leaders such as Huawei founder Ren Zhengfei and Alibaba co‑founder Jack Ma to strengthen innovation, core competitiveness, and contribute to rural revitalisation and public welfare.

Ren said China’s push to build core technologies like semiconductors and operating systems is progressing, highlighting Huawei’s HarmonyOS and its own server‑grade Arm processor, reducing reliance on foreign products.

Alibaba this week launched a series of large language models, including the new Qwen 2.5‑Max, which the company claims outperforms DeepSeek V3 and even top Western models.

Qwen 2.5‑Max, capable of generating text, video and images, achieved higher scores than OpenAI’s GPT‑4o, Anthropic’s Claude 3.5 Sonnet and Meta’s Llama 3.1 405B on benchmarks such as Arena‑Hard, MMLU‑Pro, GPQA‑Diamond, LiveCodeBench and LiveBench.

The company emphasises that its comparison is against DeepSeek V3 rather than the globally discussed R1 model, which explains why it is benchmarked against GPT‑4o instead of OpenAI’s flagship o1.

The announcement underscores that despite Western attempts to curb China’s AI progress, the United States may not be as dominant as previously thought, and the massive funding in Silicon Valley appears increasingly greedy.

Alibaba also announced that its digital infrastructure spending over the next three years will exceed the total of the past decade, though details remain vague.

China’s regulators have warned that by 2025 they will clean up locally generated AI content on the internet, reflecting concerns about AI‑generated misinformation.

Alibaba’s AI strategy aims to develop general AI, believing that human‑level AI could impact up to 50 % of global GDP.