Top AI Product

Alibaba’s Qwen team released Qwen-UI-Agent on August 20: an open-source foundation model for GUI agents. One model drives four surfaces — mobile, desktop, browser, and DeepSearch — instead of a separate agent per platform.

Trained on real phones, not simulators

Most GUI agents train in emulators and choke on real devices. Alibaba ran a live farm of 100+ physical smartphones across 150+ apps, then built MobileWorld-Real (400+ tasks, 100+ apps) to score agents on actual hardware. Results: 82.1% on MobileWorld, 92.2% on MobileWorld-Real, 79.5% on OSWorld-Verified, 81.5% on ScreenSpot-Pro — ahead of Claude Opus 4.8, Gemini 3.1 Pro, and GPT-5.6 Sol on multiple boards. The action space mixes GUI clicks with CLI commands and batches several actions per decision, so it moves faster than click-by-click agents.

Open weights you can actually run

Code, weights, and the technical report are on GitHub (Tongyi-MAI/MAI-UI), with sizes down to 2B for on-device use. Deploy locally or plug into the Qwen API ecosystem for phone automation, desktop RPA, and web tasks. Frontier labs sell computer use behind an API. Alibaba just open-sourced a version that outscores them.

You Might Also Like

• Kimi k2 6 Beats gpt 5 4 and Claude Opus 4 6 on swe Bench pro
• Bytedance ui Tars Desktop Scores 61 6 on Screenspot pro Leaving gpt 4o and Claude Behind
• Googles Gemini 3 1 Flash Live Scores 90 8 on Audio Benchmarks Real Time Voice ai Gets Serious
• Qwen 3 6 Plus vs Claude Opus 4 6 3x the Speed 1 17th the Price and the Benchmarks are Uncomfortably Close
• Hiveterm Bets on the Multi Agent Workspace Claude Codex and Gemini in one Terminal