arXiv：cs.AI· Scott C. Lowe, Sander W. Keemink·· 11 小时前
SpireBench：用《杀戮尖塔》评测智能体模型的长程策略博弈能力
SpireBench: Evaluating Long-Horizon Strategic Play of Agentic Models with Slay the Spire
AI 导读
研究者推出 SpireBench，让 20 个大语言模型在最高难度下完整游玩商业卡牌构筑游戏《杀戮尖塔》。前沿模型能击败启发式基线并追平普通人类玩家，但与专家差距明显，且随流程拉长而扩大；所有模型都会误判长程风险，倾向于保血而非以血量换取成长。全面重命名游戏术语后，模型表现部分依赖对该游戏的先验知识，且无法从跨局记录的记忆中改进。
正文
Abstract:Agentic large language models now tackle complex tasks, yet strategic reasoning over long horizons is under-studied: most agentic benchmarks are short and episodic, and performance on long tasks is hard to dissect into stored knowledge, reasoning and luck. We introduce SpireBench, in which models play the complete commercial deck-builder Slay the Spire (StS), where early decisions decide long-term success. Across 20 large language models (LLMs) at the game's highest difficulty, frontier models beat heuristic baselines and match the average human player but remain far behind experts. This gap widens as the run lengthens: the strongest models win the first act as often as experts, then fall behind. What separates strong models from weak is how well they play each fight; all of them misjudge risk over a run, preserving health rather than spending it to grow stronger. A comprehensive, self-consistent renaming of every game term shows that part of their performance rests on prior knowledge of the game, and models do not improve from the memories they write across runs, which are often wrong and overly cautious. Our harness is the first to give a model the complete original game, and we compare it with existing harnesses. Overall, SpireBench offers an unsaturated long-horizon benchmark with tools to locate where and why agents fail.
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.12596 [cs.AI] |
| (or arXiv:2610.12596v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.12596 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Scott Lowe [view email]
[v1]
Thu, 8 Oct 2026 18:18:19 UTC (2,011 KB)
来源：arXiv：cs.AI · arxiv.org