ReVision: Scaling Computer-Use Agents via Temporal Visual Redundancy Reduction

TL;DR

ReVision introduces a learned patch selector to reduce visual redundancy in graphical user interface trajectories, significantly decreasing token usage and enabling longer history processing for computer-use agents.

Contribution

It presents a novel method for visual redundancy reduction that improves efficiency and performance in multimodal language models for GUI-based agents.

Findings

ReVision reduces token usage by approximately 46% on average.

Success rate improves by 3% over the no drop baseline.

Performance continues to improve with more past observations when redundancy is removed.

Abstract

Computer-use agents (CUAs) rely on visual observations of graphical user interfaces, where each screenshot is encoded into a large number of visual tokens. As interaction trajectories grow, the token cost increases rapidly, limiting the amount of history that can be incorporated under fixed context and compute budgets. This has resulted in no or very limited improvement in the performance when using history unlike other domains. We address this inefficiency by introducing ReVision, which is used to train multimodal language models on trajectories where redundant visual patches are removed using a learned patch selector that compares patch representations across consecutive screenshots while preserving spatial structure required by the model. Across three benchmarks, OSWorld, WebTailBench, and AgentNetBench, when processing trajectories with 5 history screenshots using Qwen2.5-VL-7B, ReVision reduces token usage by approximately 46% on average while improving success rate by 3% over the no drop baseline. This establishes a clear efficiency gain, enabling agents to process longer trajectories with fewer tokens. With this improved efficiency, we revisit the role of history in CUAs and find that performance continues to improve as more past observations are incorporated when redundancy is removed. This suggests that the commonly observed saturation in visual history is not due to limited usefulness of past information, but rather a consequence of inefficient token representations.