OmniGUI: Benchmarking GUI Agents in Omni-Modal Smartphone Environments

2026-04-03

arXiv.org

AI Summary

This study addresses the limitation of existing GUI agent benchmarks that rely on static screenshots, which fail to evaluate agents' capacity for dynamic interactions involving real-time audio and video. To this end, we construct the first omni-modal, step-level smartphone benchmark providing continuously interleaved images alongside synchronized audio and video inputs, annotated with multimodal dependency levels. Methodologically, we employ a base omni-modal model natively supporting interleaved inputs as the agent, conducting ablation studies using expert demonstration data. Our findings reveal that current models suffer significant performance degradation when synchronous spatiotemporal signals are required, identifying cross-modal interference as the primary bottleneck.

Abstract

Current benchmarks for graphical user interface (GUI) agents predominantly rely on static screenshots. However, real-world smartphone interaction routinely requires agents to process transient audio cues and temporal video dynamics that are tightly coupled with the moment of action. To bridge this gap, we introduce OmniGUI, the first step-level benchmark designed to evaluate GUI agents in omni-modal smartphone environments. OmniGUI provides continuous, interleaved multimodal inputs comprising static images, synchronous audio, and video clips at every action step. The dataset encompasses 709 expert-demonstrated episodes (2,579 action steps) across 29 applications, systematically annotated with objective multimodal dependency levels. Because dedicated omni-modal GUI agent frameworks are currently in their nascent stage, we select foundational omni-modal models capable of natively processing interleaved inputs to serve as agent proxies for our initial baselines. Our empirical evaluation reveals that while current models exhibit competency on visually static tasks, their action prediction performance degrades significantly in environments requiring synchronous temporal and auditory signals. Furthermore, ablation studies isolate specific operational bottlenecks, notably cross-modal interference when processing task-irrelevant environmental noise. The complete dataset, evaluation pipeline, and baseline prompts are provided in the supplementary material. Project page: https://omni-gui.github.io.

Problem

• GUI agents
• omni-modal benchmark
• smartphone environments
• multimodal interaction
• action prediction

Innovation

• Omni-modal GUI agents
• Step-level benchmark
• Interleaved multimodal inputs
• Cross-modal interference
• Smartphone environments