@inproceedings{wu-etal-2026-os,
title = "{OS}-Marathon: Benchmarking Computer-Use Agents on Vast-Horizon, Repetitive Tasks",
author = "Wu, Jing and
Ai, Wenjie and
Barretto, Daphne and
Chen, Yiye and
Chen, Qingyu and
He, Yuhang and
Chawla, Pranit and
Gyd{\'e}, Nicholas and
Jian, Yanan and
Vineet, Vibhav",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2493/",
pages = "56604--56622",
ISBN = "979-8-89176-459-0",
abstract = "Vast-horizon, repetitive workflows are common in daily routines, e.g., processing expense reports from a stack of receipts, organising a collection of PDF annotations into structured notes, and are tedious for humans, with execution length scaling with the volume of data to process. They are, however, well-suited to autonomous agents, since their structured, recurring sub-workflows follow a logic that can be systematically learned. Yet no existing benchmark evaluates agents specifically on vast-horizon repetitive capabilities, leaving the problem largely underexplored. To bridge this gap, we establish $\text{OS-Marathon}$, comprising 100 vast-horizon repetitive tasks across 5 scenarios and 10 domains, on which our evaluation reveals that leading state-of-the-art (SOTA) CUAs struggle substantially. We further find that introducing a task orchestrator to decompose the workflow into per-instance subtasks fails to mitigate the difficulty: errors accumulate and propagate across solver agents, indicating that naive decomposition is insufficient for these tasks. We then explore a cost-effective personalisation strategy, i.e. GraphDemo, that adapts general agents to such tasks from a single human demonstration of the recurring sub-workflow logic. Extensive experiments show both the real-world challenge of vast-horizon repetitive tasks and the improvement provided by human demonstration in this setting."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wu-etal-2026-os">
<titleInfo>
<title>OS-Marathon: Benchmarking Computer-Use Agents on Vast-Horizon, Repetitive Tasks</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jing</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wenjie</namePart>
<namePart type="family">Ai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Daphne</namePart>
<namePart type="family">Barretto</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yiye</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qingyu</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuhang</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pranit</namePart>
<namePart type="family">Chawla</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nicholas</namePart>
<namePart type="family">Gydé</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanan</namePart>
<namePart type="family">Jian</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vibhav</namePart>
<namePart type="family">Vineet</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Vast-horizon, repetitive workflows are common in daily routines, e.g., processing expense reports from a stack of receipts, organising a collection of PDF annotations into structured notes, and are tedious for humans, with execution length scaling with the volume of data to process. They are, however, well-suited to autonomous agents, since their structured, recurring sub-workflows follow a logic that can be systematically learned. Yet no existing benchmark evaluates agents specifically on vast-horizon repetitive capabilities, leaving the problem largely underexplored. To bridge this gap, we establish \textOS-Marathon, comprising 100 vast-horizon repetitive tasks across 5 scenarios and 10 domains, on which our evaluation reveals that leading state-of-the-art (SOTA) CUAs struggle substantially. We further find that introducing a task orchestrator to decompose the workflow into per-instance subtasks fails to mitigate the difficulty: errors accumulate and propagate across solver agents, indicating that naive decomposition is insufficient for these tasks. We then explore a cost-effective personalisation strategy, i.e. GraphDemo, that adapts general agents to such tasks from a single human demonstration of the recurring sub-workflow logic. Extensive experiments show both the real-world challenge of vast-horizon repetitive tasks and the improvement provided by human demonstration in this setting.</abstract>
<identifier type="citekey">wu-etal-2026-os</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2493/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>56604</start>
<end>56622</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T OS-Marathon: Benchmarking Computer-Use Agents on Vast-Horizon, Repetitive Tasks
%A Wu, Jing
%A Ai, Wenjie
%A Barretto, Daphne
%A Chen, Yiye
%A Chen, Qingyu
%A He, Yuhang
%A Chawla, Pranit
%A Gydé, Nicholas
%A Jian, Yanan
%A Vineet, Vibhav
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F wu-etal-2026-os
%X Vast-horizon, repetitive workflows are common in daily routines, e.g., processing expense reports from a stack of receipts, organising a collection of PDF annotations into structured notes, and are tedious for humans, with execution length scaling with the volume of data to process. They are, however, well-suited to autonomous agents, since their structured, recurring sub-workflows follow a logic that can be systematically learned. Yet no existing benchmark evaluates agents specifically on vast-horizon repetitive capabilities, leaving the problem largely underexplored. To bridge this gap, we establish \textOS-Marathon, comprising 100 vast-horizon repetitive tasks across 5 scenarios and 10 domains, on which our evaluation reveals that leading state-of-the-art (SOTA) CUAs struggle substantially. We further find that introducing a task orchestrator to decompose the workflow into per-instance subtasks fails to mitigate the difficulty: errors accumulate and propagate across solver agents, indicating that naive decomposition is insufficient for these tasks. We then explore a cost-effective personalisation strategy, i.e. GraphDemo, that adapts general agents to such tasks from a single human demonstration of the recurring sub-workflow logic. Extensive experiments show both the real-world challenge of vast-horizon repetitive tasks and the improvement provided by human demonstration in this setting.
%U https://aclanthology.org/2026.emnlp-main.2493/
%P 56604-56622
Markdown (Informal)
[OS-Marathon: Benchmarking Computer-Use Agents on Vast-Horizon, Repetitive Tasks](https://aclanthology.org/2026.emnlp-main.2493/) (Wu et al., EMNLP 2026)
ACL
- Jing Wu, Wenjie Ai, Daphne Barretto, Yiye Chen, Qingyu Chen, Yuhang He, Pranit Chawla, Nicholas Gydé, Yanan Jian, and Vibhav Vineet. 2026. OS-Marathon: Benchmarking Computer-Use Agents on Vast-Horizon, Repetitive Tasks. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 56604–56622, Budapest, Hungary. Association for Computational Linguistics.