AutoGUI-v2 - Region Grounding (Functionality): leaderboard

AutoGUI-v2 (2026) tests functionality understanding of GUI regions and elements on web, desktop and mobile screenshots. This board asks the model to locate a functional region from what it does; a prediction counts when its point falls inside the ground-truth box. Accuracy in percent; higher is better.

Metric: Region grounding by functionality on 442 tasks, grounding center accuracy (%): the predicted point falls inside the ground-truth box; native resolution, the authors' tuned prompts; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GLM-4.5V84.6#754
2Qwen 3 VL 32B Instruct84.4#606
3Gemini 2.5 Pro77.6#276
4uitars-1.5-7B75.8#7905
5Qwen 3 VL 8B Instruct70.1#1029
6UGround-V1-7B64.7
7OS-Atlas-Base-7B56.3
8Claude Sonnet 4.548.6#261
9O336.7#237
10Step-336.7#734
11GPT-531.4#178

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=autogui-v2-region-grounding-functionality · How It Works · Data refreshed daily, snapshot 2026-11-21.