Alibaba’s Qwen Office Beta Launches as AI Office Battle Heats Up; Chinese Models Dominate Multimodal Rankings
Alibaba’s Qwen Office enters beta, integrating three AI agents to compete with Tencent WorkBuddy. Meanwhile, Chinese models like Zhiyuan, Qwen, and...
Alibaba has officially launched the beta version of its AI office product "Qwen Office" on July 27, marking a key move in the intensifying AI office competition among tech giants. The product integrates three internal agent tools—QoderWork, Wukong, and MuleRun—under the leadership of DingTalk CEO Chen Yusen, positioning it as Alibaba’s top strategic project for its to B business.
Qwen Office, built on the Tongyi Qwen large model, offers a desktop client (Windows and macOS Beta) with plans to roll out web and DingTalk-integrated versions. It uses the Qwen3.8-Max-preview model, supporting million-level context and code generation, with subscription options for free, personal, and enterprise users. Alibaba aims to make it its flagship SaaS product for the second half of 2026, with pricing slightly lower than Tencent’s WorkBuddy and integration with ecosystems like Taobao and Cainiao.
The AI office space is becoming a fierce battleground. Tencent’s WorkBuddy leads the desktop AI agent market with over 20 million monthly PC visits in June. ByteDance is merging Feishu with Doubao to focus on workflow optimization, while Alibaba, Tencent, and ByteDance are all vying for dominance in full-process desktop operation capabilities. Industry insiders view AI office as the most validated monetization path for large models, where enterprise payment willingness will determine future market rankings.
In user metrics, Doubao’s monthly active users (MAU) reached a record 528 million in June. Qwen and Kuaike have stabilized at over 200 million MAU each, with Qwen topping media buzz for three consecutive months since April—ending Doubao’s nine-month lead.
Chinese AI models also shone in multimodal rankings. On July 28, Fudan University released the DailyOmni audio-video cross-modal reasoning list, where Zhiyuan’s Silicon Photonics Dynamic Language Model took the top spot with 85.21 points, outperforming Qwen, Google Gemini, Doubao, and NVIDIA models. Qwen and Doubao were among the top five, highlighting China’s strengths in multimodal AI perception.
Additionally, Qwen3.8-Max-Preview, released on July 19, gained traction overseas. Touted as "second only to Fable 5," the model went viral on social media for its ability to generate usable web and mobile apps from requirements without prior documentation, at a fraction of the cost of overseas models. Alibaba has announced plans to open-source the model soon.
Sources