What it measures
Evaluates visual UI element grounding (clicking exact bounding boxes from natural-language instructions across mobile, desktop, and web).
Reading the measurement
The useful outcome is a verified change or deliverable. Inspect the tool interface, initial state, user simulator and policy constraints. An agent can claim completion without reaching the required state, or reach it through an action the protocol forbids.
Published evaluation
The source average covers screen grounding across mobile, desktop and web. ScreenSpot and ScreenSpot-v2 stay separate measures.
Where this fits in an evaluation suite
A transcript that says “done” is weaker evidence than the verified final state. Track policy violations, recoveries, and invalid attempts alongside success.
Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.
A reproducible comparison
The source average covers screen grounding across mobile, desktop and web. ScreenSpot and ScreenSpot-v2 stay separate measures.
Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.
PRIMARY EVIDENCE & FURTHER READING
- OS-ATLAS: A Foundation Action Model for Generalist GUI Agents — published result table ↗
All 11 authors
Zhiyong Wu, Zhenyu Wu, Fangzhi Xu, Yian Wang, Qiushi Sun, Chengyou Jia, Kanzhi Cheng, Zichen Ding, Liheng Chen, Paul Pu Liang, Yu Qiao
Reviewed 2026-10-26. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.