Saved in:
Bibliographic Details
Main Authors: Xie, Jingxu, Xu, Dylan, Zhao, Xuandong, Song, Dawn
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2506.14205
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910036206288896
author Xie, Jingxu
Xu, Dylan
Zhao, Xuandong
Song, Dawn
author_facet Xie, Jingxu
Xu, Dylan
Zhao, Xuandong
Song, Dawn
contents We introduce AgentSynth, a scalable and cost-efficient pipeline for automatically synthesizing high-quality tasks and trajectory datasets for generalist computer-use agents. Leveraging information asymmetry, AgentSynth constructs subtasks that are simple during generation but significantly more challenging when composed into long-horizon tasks, enabling the creation of over 6,000 diverse and realistic tasks. A key strength of AgentSynth is its ability to precisely modulate task complexity by varying the number of subtasks. Empirical evaluations show that state-of-the-art LLM agents suffer a steep performance drop, from 18% success at difficulty level 1 to just 4% at level 6, highlighting the benchmark's difficulty and discriminative power. Moreover, our pipeline achieves a low average cost of $0.60 per trajectory, orders of magnitude cheaper than human annotations. Our code and data are available at https://github.com/sunblaze-ucb/AgentSynth
format Preprint
id arxiv_https___arxiv_org_abs_2506_14205
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle AgentSynth: Scalable Task Generation for Generalist Computer-Use Agents
Xie, Jingxu
Xu, Dylan
Zhao, Xuandong
Song, Dawn
Computation and Language
We introduce AgentSynth, a scalable and cost-efficient pipeline for automatically synthesizing high-quality tasks and trajectory datasets for generalist computer-use agents. Leveraging information asymmetry, AgentSynth constructs subtasks that are simple during generation but significantly more challenging when composed into long-horizon tasks, enabling the creation of over 6,000 diverse and realistic tasks. A key strength of AgentSynth is its ability to precisely modulate task complexity by varying the number of subtasks. Empirical evaluations show that state-of-the-art LLM agents suffer a steep performance drop, from 18% success at difficulty level 1 to just 4% at level 6, highlighting the benchmark's difficulty and discriminative power. Moreover, our pipeline achieves a low average cost of $0.60 per trajectory, orders of magnitude cheaper than human annotations. Our code and data are available at https://github.com/sunblaze-ucb/AgentSynth
title AgentSynth: Scalable Task Generation for Generalist Computer-Use Agents
topic Computation and Language
url https://arxiv.org/abs/2506.14205