Qwen Office Pulls Third-Party AI Models: Tension Between Openness and Core Model Protection

Qwen Office, Alibaba's AI-powered office tool, recently made a sudden move that caught users off guard: on the evening of August 17, it removed two third-party cutting-edge models—Zhipu GLM-5.3 and DeepSeek V4 Pro—from its model options. The models vanished without any official announcement, and even the public account article that once promoted these "cutting-edge models" was set to private, making it inaccessible to users. Adding to the confusion, a prior announcement promising unlimited access to GLM-5.3 from August 17 to 18 was quickly deleted, leading netizens to joke that the platform's operational changes were faster than model iterations.

This shift comes after Qwen Office had rapidly integrated the two models just days earlier, on August 14, right after their release. The move was seen as a demonstration of Qwen Office's support for multi-model aggregation—a strategy that allows users to complete various tasks on a single platform without worrying about which model is handling the job, thus boosting task completion rates.

However, behind this adjustment lies an internal conflict at Alibaba: the product team and the model team have differing priorities. The product team aimed to enhance user appeal by integrating more external models to enrich features, while the model team focused on promoting Qwen's proprietary models to collect real-world task data, which is crucial for iterating and improving their own AI models. This clash highlights the broader industry tension between maintaining an open ecosystem and protecting core proprietary technologies.

The industry context amplifies this dilemma. Major players like Tencent, ByteDance, and Baidu are all exploring model aggregation, but with limited openness—each platform operates within its own "ecosystem wall" (e.g., WorkBuddy doesn't include Qwen or Seed models, while TRAE Work excludes Hunyuan). Additionally, DeepSeek's recent price hike (up to 1100% increase) has fueled concerns among users that the era of low-cost domestic AI tools might be coming to an end. For workers who had grown accustomed to switching between models for complex tasks, Qwen Office's removal of third-party options left many feeling disoriented.

Model aggregation, while beneficial for users, poses risks for platforms. If users prefer external models, the platform's own model may receive less usage, leading to a lack of data for improvement—a classic "tragedy of the commons" scenario. Hence, most platforms opt for selective openness, balancing user needs with the growth of their proprietary models.

Sources

• Toutiao Article: https://m.toutiao.com/article/7675979163090321972/