Metraipolis › archive › dc_bdba1ec7696a
LongPuzzleBench: Evaluating GUI Agents on Long-Horizon Visual Puzzles — arXiv:2609.34769
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.34769
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T21:02:56.498Z
- SHA-256 of the archived text
89446fe0c88ff9fefc208a961ab87175473e7218f9d760bad50454843f1269b7- Stable URL of this page
https://metraipolis.com/documento/dc_bdba1ec7696a
How to cite
LongPuzzleBench: Evaluating GUI Agents on Long-Horizon Visual Puzzles — arXiv:2609.34769 — arXiv (en), https://arxiv.org/abs/2609.34769 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_bdba1ec7696a (Metraipolis archive dc_bdba1ec7696a, sha256 89446fe0c88ff9fefc208a961ab87175473e7218f9d760bad50454843f1269b7)
BibTeX
@misc{dcbdba1ec7696a,
title = {LongPuzzleBench: Evaluating GUI Agents on Long-Horizon Visual Puzzles — arXiv:2609.34769},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.34769},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_bdba1ec7696a; sha256 89446fe0c88ff9fefc208a961ab87175473e7218f9d760bad50454843f1269b7},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.34769. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
GUI agents need long-horizon visual reasoning: they must interpret a changing interface while keeping a multi-step plan viable as earlier actions constrain later ones. Existing benchmarks evaluate grounding, computer use, and game play, but rarely test whether agents stay coherent across long chains of coupled decisions. Long-horizon visual puzzles expose this capability directly: a legal move that looks like progress can make the puzzle unsolvable, and the loss shows only several moves later. We introduce LongPuzzleBench, 114 levels in six puzzle games played through native GUI actions, where one objective can take a human over a thousand actions on persistent boards and dead ends go unannounced. With Native GUI Actions alone, the strongest agents solve most objectives, but success falls sharply on harder, longer boards: seven of ten general-purpose agents solve nothing harder than Medium, and none completes Bolt Unscrew Hard, which a human solves along with every other objective. Code Execution CUA does not close this gap, and its scores mix visual solving with algorithmic search. Controlled diagnostics trace these failures to one limitation that neither rules, state hints, nor failure memory removes: agents judge each move by the visible progress it makes, not by the future options it leaves. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.34769 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T20:29:03.155Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).