Chinese AI Models Sweep Top Two Slots on ScreenSpot-Pro

On September 9, 2026, ScreenSpot-Pro, a globally recognized GUI grounding benchmark, updated its leaderboard. Indeed-UI-32B, developed by a Hangzhou-based Chinese AI company, topped the global ranking with an 82.7% task success rate, while Indeed-UI-8B followed at 81%, securing second place. The two models claimed the top two positions, ahead of leading global models such as GPT, Claude, Gemini, and Qwen.

ScreenSpot-Pro has been included in Hugging Face's official benchmark system and is widely regarded as an industry standard for evaluating GUI grounding capabilities. The benchmark uses full-screen, high-resolution screenshots of 26 real professional applications across Windows, macOS, and Linux, testing whether AI can accurately locate buttons, menus, or input boxes in complex software interfaces.

A representative from the company said the achievement demonstrates leadership in both absolute performance and efficiency, achieving high precision with smaller parameter sizes and providing a practical path for low-cost enterprise deployment. The company previously ranked first on OSWorld, another internationally recognized benchmark. Moving forward, the Indeed-UI series will continue to iterate in challenging areas such as icon localization, complex nested interfaces, and cross-resolution scenarios.