AI summaryⓘ
The authors studied how running extra computations during use (inference-time scaling) affects computer-use agents (CUAs) working on limited hardware, like personal devices. They found that while giving the agents more historical context helps them be more stable and accurate, the improvements eventually level off and cause new types of errors. Extending how long the agents plan doesn’t greatly increase success and might just prolong mistakes. Breaking down tasks into parts or running things in parallel can help but often adds overhead or costs more computing power. Overall, the authors suggest that making these agents efficient on local machines means carefully choosing where to spend computing resources and building systems aware of their limitations.
autonomous agentsinference-time scalingcontextual scalingtemporal scalingstructural decompositionparallel scalinglocal modelsOSWorld benchmarkcomputational costtask accuracy
Abstract
Deploying autonomous computer-use agents (CUAs) locally is increasingly important for privacy, cost efficiency, and practical usability, yet improving their performance under strict hardware constraints remains challenging. While recent studies show that inference-time scaling can improve frontier computer-use agents through additional computation during execution, its effectiveness for resource-constrained local models remains poorly understood. We present a systematic empirical study of inference-time scaling in local CUAs across contextual, temporal, structural, and parallel dimensions. We evaluate Qwen3-VL-8B/30B-A3B, UI-TARS-1.5-7B, and OpenCUA-7B on the OSWorld benchmark. Our results show that additional computation often yields diminishing returns while changing failure modes. Contextual scaling provides historical grounding that improves trajectory stability and task accuracy, but its gains saturate as token cost increases and failures shift from repetitive or stalled trajectories toward premature false successes. Temporal scaling similarly reduces max-step stalls, yet does not substantially improve task success, indicating that longer horizons often extend erroneous trajectories rather than correct them. We further find that structural decomposition can introduce planning and formatting overhead in local two-stage agents, while parallel scaling partially mitigates these failures at a substantial computational cost. Overall, our findings suggest that efficient local CUAs require selective compute allocation, failure-aware control mechanisms, and agentic frameworks designed around the capabilities and limitations of local models.