Alibaba releases a new large model that can run on laptops, directly confronting Meta's AI offensive

Key Points

1. Alibaba has launched the Tongyi new AI model, specifically designed to run on consumer-grade hardware such as laptops.
2. At the same time, it has opened the model weights of its strongest model, Qwen3.8 Max.
3. This move further intensifies the competition between Alibaba and Meta in the open-source weight AI track.

Meta aims to lead the open-source AI track, while Chinese tech giant Alibaba has just given this American tech company a lesson, demonstrating how fierce the competition in this field is.

On Monday, this Chinese tech company released an AI large model aimed at consumer terminal hardware such as laptops, marking the terminal-side model as a new battleground for large model vendors; at the same time, Alibaba opened the weight files of its strongest model, striving to maintain its leading position in the open-source technology field.

Just last week, Meta announced plans to open-source its strongest AI large model while launching several new models that can run locally on laptops. Meta hopes to create an American AI solution to compete with closed-source top laboratories like OpenAI and Anthropic, countering Chinese AI technology.

The Qwen3.8‑27B released by Alibaba excels in code writing, professional office tasks, scientific research, and long-cycle intelligent agent tasks, with officials claiming its performance can reach the level of a large model ten times its own scale.

Alibaba also opened the weight files of its flagship model Qwen3.8 Max. The weights contain the computational logic and behavioral rules for running the AI model, and opening the weights means developers can freely download and deploy it, but the data and training methods used to train the model have not been disclosed.

Alibaba has become a leader in the open-source weight AI field, with domestic companies like DeepSearch and Moonlight also showing strong capabilities.

Meta, with its Llama series, was an early entrant in open-source AI but was quickly surpassed by Chinese manufacturers. Last week, Meta released the Muse Glimmer open-source model series, focusing on local operation on laptops.

Nick Pesce, AI head of Futurum Group, stated in an interview: "Meta's renewed investment in the open-source weight track is itself a response to the significant market share that Chinese laboratories have captured over the past two years."

Alibaba Strives to Consolidate Its Dominance

The success of open-source weight models largely depends on download volumes and the number of developers engaged in secondary development based on the model.

Data from the world's largest open-source model repository, Hugging Face, showed last week that the number of models derived from Tongyi (Qwen) has reached 151,448, which is 2.6 times the scale of all Meta's model ecosystem Neil Shah, co-founder of Counterpoint Research, stated: "Whoever can provide the most powerful open-source weight model will gain an advantage in this competition."

"Alibaba aims to become the undisputed industry leader, surpassing Meta, targeting the global market, and becoming a strong alternative to the top deployable large models in Silicon Valley."

The release of Qwen 3.8-27B indicates that Alibaba believes high-performance large models will increasingly run on the edge rather than just in data centers. Industry experts believe that edge AI running on local hardware in smartphones and laptops is faster and more secure.

Shah noted that the "next battlefield" for large models is local deployment on the edge, rather than relying on cloud data centers. Pei Shen stated that Alibaba has already established advantages in multiple dimensions of open-source weights and edge AI.

"Tongyi (Qwen) has become the most compelling non-U.S. model system, with high recognition both in local hardware collaborations in China and among the global open-source developer community," Pei Shen said