Alibaba's new AI model Qwen surpasses Meta's Llama and DeepSeek

Alibaba’s Qwen open-source AI model is gaining momentum globally, accounting for more than 50% of open-source AI model downloads, according to SCMP. Research by Interconnects AI shows that by March 2026 Qwen had accumulated nearly 1 billion cumulative downloads, outpacing major rivals including Meta Platforms’ Llama and models from DeepSeek. In February alone, Qwen recorded 153.6 million downloads—more than twice the combined total of the next eight major rival models—based on data compiled from Hugging Face.

Download momentum and open-source expansion

Analysts say Qwen’s lead in download counts is helping Alibaba rapidly expand its developer community, a key advantage in the open-source AI race. The scale of adoption also reinforces Qwen’s visibility among developers and organizations building on open-source models.

Alibaba’s push beyond chat: “world model” ambitions

The US-China AI race is intensifying as Alibaba broadens its strategy. On April 9, CNBC reported that Alibaba is expanding into a new direction: a “world model” intended to simulate and understand the real world using multimodal data such as video, audio, and physical interaction.

Unlike text-first chatbots such as ChatGPT, the world model concept emphasizes “perception and action,” aiming to help AI predict and interact with physical environments. The approach is positioned as a foundation for technologies including robots, autonomous vehicles, and automation systems.

Investment in ShengShu and Vidu’s positioning

To pursue this ambition, Alibaba reportedly led an investment of roughly CNY 2 billion (about USD 290 million) into ShengShu, the startup behind the AI video tool Vidu. The company’s latest product, Vidu Q3 Pro, is currently among the top 10 AI video generation models worldwide.

Vidu also debuted globally before OpenAI’s widely adopted video tool Sora, according to the report.

China’s broader AI ecosystem

Beyond Alibaba, China is building a competitive ecosystem of AI offerings, including Baidu’s Ernie Bot, Tencent’s Hunyuan AI, and ByteDance’s Doubao AI. Together, these efforts reflect a push to scale AI capabilities across consumer and enterprise use cases.

US incumbents retain advantages in deployment

Despite China’s rapid progress, US tech giants still hold a strong lead in ecosystem breadth and practical applications. OpenAI’s ChatGPT remains a leading consumer AI product, with deep integration into business and education. Microsoft is also developing Copilot, bringing AI into office tools such as Word, Excel, and Windows.

NVIDIA does not directly compete in consumer-scale language models, but it dominates AI infrastructure through CUDA and GPU optimization models. Meta Platforms’ Llama continues to be important in the open-source community, particularly for researchers and startups.

Implications for the AI race

Overall, the developments suggest the AI race is no longer a Silicon Valley-only competition. Instead, it is evolving into a faster-paced US-China contest, with open-source adoption and multimodal “world model” ambitions shaping the next phase of competition.

Sources cited: CNBC and SCMP.