Sign InOpen Brain
arXivPaperNeeds Review

Desktop-Delta Bench: Do Computer-Use Models Understand Desktop GUI Transitions?

Desktop-Delta Bench tests whether computer-use models can verify what changed after an action, exposing a reliability gap hidden by end-task and single-frame scores.

arXiv · Jul 28, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Desktop-Delta Bench contains **2,013 human-verified instances** from multi-app Linux trajectories across about 15 applications and 50 task domains. It tests temporal ordering and action inference from before-and-after observations.

Practical Implication

Use step-level transition checks when evaluating computer-use agents, especially around stale screenshots, delayed rendering, and recovery. The best ordering exact-match rates reached only **65.1% without decoys** and **65.7% with decoys**.

Agent-Ready Context
Desktop-Delta Bench contains **2,013 human-verified instances** from multi-app Linux trajectories across about 15 applications and 50 task domains. It tests temporal ordering and action inference from before-and-after observations.

Use step-level transition checks when evaluating computer-use agents, especially around stale screenshots, delayed rendering, and recovery. The best ordering exact-match rates reached only **65.1% without decoys** and **65.7% with decoys**.

This is an offline diagnostic rather than an end-to-end task benchmark. Action recognition also varies sharply: reported click and drag scores were **0.96 versus 0.76 F1**, so aggregate GUI grounding can hide operation-specific weaknesses.
Connected Context · Feed7 Judgment

Desktop-Delta Bench narrows the evaluation gap identified by broader coding and rollout benchmarks to a specific failure surface: whether an agent understands the transition between GUI states. Its low ordering ceiling and operation-specific score spread support adding step-level temporal and action checks to reproducible trajectory evaluation, rather than treating end-to-end completion or aggregate grounding as sufficient.

Everything Is a Rollout — Alex Shaw + Ryan Marten, Terminal-Bench, Harbor, Laude InstituteHarbor’s trajectory inspection provides the evaluation loop into which Desktop-Delta’s step-level transition checks can fit.From Agent Traces to Agent Simulations — Rustem Feyzkhanov, Snorkel AIReplayable production traces could supply fixed GUI transitions for diagnosing the stale-state, rendering-delay, and recovery failures this benchmark isolates.DeepSWE: A Contamination-Resistant Coding Benchmark — James Shi, DatacurveBoth show that end scores can conceal behavior-level weaknesses, but Desktop-Delta extends trajectory-sensitive evaluation from repository work to desktop state transitions.How Evals and Prompts Shape Agent Behavior — Preetika Bhateja & Daniel Bump, YouTube AdsDesktop-Delta supplies the small, behavior-specific transition cases that can make trace review and calibrated evaluation more diagnostic for computer-use agents.
Context Map
benchmarkcoding#computer-use#agent-evals#agent-reliability
Uncertainty
This is an offline diagnostic rather than an end-to-end task benchmark. Action recognition also varies sharply: reported click and drag scores were **0.96 versus 0.76 F1**, so aggregate GUI grounding can hide operation-specific weaknesses.