88Z AI, 5 August — Alibaba's Qwen team and XLang Lab have jointly released Qwen-CUA, a native computer use agent. Built on a mixture-of-experts architecture, it perceives interfaces via screenshots and directly outputs keyboard and mouse events to control software without relying on underlying metadata. It performs well on the OSWorld-Verified benchmark, with the trillion-parameter version scoring 87.6.
At a glance
- Qwen-CUA perceives interfaces via screenshots and directly outputs input events
- The trillion-parameter Max version scored 87.6 on the OSWorld-Verified benchmark
The update
Alibaba's Qwen team, in collaboration with XLang Lab, has officially launched Qwen-CUA, a native computer use agent. Utilizing a 397B-A17B mixture-of-experts architecture, the model innovatively perceives interface states through screenshots, eliminating reliance on DOM trees or accessibility metadata. It directly generates keyboard and mouse events to enable precise control over browsers, desktop environments, and professional software. In performance evaluations, Qwen-CUA achieved a score of 86.2 on the OSWorld-Verified benchmark. Its trillion-parameter Max version performed even better, reaching a score of 87.6, demonstrating robust automation capabilities.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- Qwen-CUASource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.