LongPuzzleBench: Evaluating GUI Agents on Long-Horizon Visual Puzzles

LongPuzzleBench evaluates coherent long-horizon planning on persistent visual puzzle boards through native GUI actions. It compares GUI-only and code-assisted agents and uses controlled hints and memory diagnostics to examine decisions that preserve or eliminate future solutions.

AI-assisted editorial review against a primary source. This is a catalog annotation, not an independent reproduction or endorsement.

What is this resource?

Resource type

Research papers

Environment

unspecified

Topics

evaluation, planning, long-horizon, state-tracking

Record publication date

2026-09-28

Evidence retrieved

2026-10-16T06:36:29Z

Resource ID

arxiv:2609.34769

Recorded paper version

arXiv v1

Version updated: 2026-09-28T09:49:25Z.

This identifies the retained arXiv metadata snapshot; newly observed changes require separate review.

Authors

Bingo Zhang, Haochuan Lu, Zongjie Li, Genjian Li, Ari Yu Zhang, Chaozheng Wang

Cite this paper

Download BibTeX · Download CSL JSON

Citations preserve the selected source edition, its authors and recorded date precision. Unrecorded venues and identifiers are omitted.

Explore related topics

Scope and limitations

The annotation and scope were reviewed against the complete retained original arXiv abstract. Experiments, reported performance and artifact availability have not been independently reproduced. Agents execute real GUI moves in six puzzle games; the tasks diagnose computer-control planning but remain game-based rather than general productivity workflows.

Primary evidence and verification

Source retrieval is separate from experiment reproduction. Use the original publication or publisher page when citing claims about the resource.

1. arxiv api

   Retrieved 2026-10-16T06:36:29Z. Supports: name, authors, published_at, updated_at, identifiers.arxiv, details.arxiv_snapshot.

   Content fingerprint

   74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f
2. editorial primary review

   Retrieved 2026-10-16T06:36:29Z. Supports: name, kind, summary, environments, topics, scope, notes.

   Content fingerprint

   74e521e46c46489e48f9bad5d774625a85b5ccd32828969fd98e1a4c453b899f

Suggest a correction

Open a source-backed correction. Include this record ID and the original source. Read our editorial policy and coverage boundaries.