Qwen-CUA: Native Computer Use for (almost) Everything

2026-08-03

AI Summary

This work addresses the challenge of building general-purpose software agents capable of completing tasks using only screenshots and keyboard-mouse inputs, without relying on APIs or structured metadata. Key difficulties include long-horizon state tracking, sparse rewards, and the need for large-scale interaction data. To tackle these, we introduce a native computer usage framework built upon the Qwen Mixture-of-Experts large language model (up to one trillion parameters), integrating multi-screenshot state representations, fixed-chunk history compression, and joint supervised and reinforcement learning from trajectory slices. Leveraging cloud-based clusters, we collect tens of thousands of verifiable tasks for training and evaluation. Our approach achieves near-human performance in general software interaction without DOM or accessibility interfaces, scoring 87.6 on OSWorld-Verified, with binary and partial completion rates of 21.2% and 53.3%, respectively, and reduces RedTeamCUA attack success rates to 16.4%.

Abstract

Native computer use offers a general interface for agents to operate almost any software available to people, but requires long-horizon state tracking, large-scale interactive experience, and learning from sparse yet verifiable outcomes. We introduce Qwen-CUA, a native computer-use agent with a 397B-A17B Qwen mixture-of-experts backbone. It observes only screenshots and acts through keyboard and mouse events, without DOM trees, accessibility metadata, or task-specific APIs. Its scaffold maintains up to 20 active screenshots and folds older visual history in fixed-size blocks to retain recent evidence while preserving reusable prompt prefixes. For training, we build a cloud rollout fleet with access to nearly 100,000 vCPUs and tens of thousands of concurrent environments, construct approximately 40,000 verifiable tasks, and collect personalized long-horizon workflows across everyday and professional software. We optimize complete trajectories with verifiable rewards and trajectory slicing, while iterative training runs refresh supervised data and recalibrate reinforcement-learning tasks. Across eight benchmarks, Qwen-CUA outperforms Qwen3.7 and remains competitive with leading proprietary systems, reaching 86.2 on OSWorld-Verified and 18.5/48.4 binary/partial completion on OSWorld 2.0. Scaling the same recipe to a model with over one trillion parameters yields Qwen-CUA-Max, improving these scores to 87.6 and 21.2/53.3. Qwen-CUA also reduces RedTeamCUA attack success from 36.6 to 16.4 relative to Qwen3.7. Efficiency analyses, a browser deployment, and Bash-augmented experiments further characterize practical behavior. These results establish native computer use as a broadly capable agent foundation and highlight scalable verifiable interaction and hybrid tool use as key directions.

Problem

• native computer use
• long-horizon state tracking
• sparse rewards
• verifiable tasks
• general-purpose agent

Innovation

• native computer use
• mixture-of-experts
• verifiable interaction
• long-horizon agent
• scalable reinforcement learning