OdysseyArena：面向长时程、主动与归纳式交互的大语言模型基准测试
OdysseyArena: Benchmarking Large Language Models For Long-Horizon, Active and Inductive Interactions
研究者推出 OdysseyArena，将 LLM 智能体评测重心转向长时程、主动与归纳式交互，并形式化实例化四种原语。其中 OdysseyArena-Lite 提供 120 项任务衡量归纳效率与长时程发现能力，OdysseyArena-Challenge 则在超过 200 步的极端交互时程下压测稳定性。对 15+ 个领先 LLM 的实验显示，即便前沿模型在归纳场景中仍存在明显不足。
Authors:Fangzhi Xu, Hang Yan, Qiushi Sun, Jinyang Wu, Zixian Huang, Muye Huang, Jingyang Gong, Zichen Ding, Kanzhi Cheng, Yian Wang, Xinyu Che, Zeyi Sun, Jian Zhang, Zhangyue Yin, Haoran Luo, Ben Kao, Qika Lin
Abstract:The advancement of Large Language Models (LLMs) has catalyzed the development of autonomous agents capable of navigating complex environments. However, existing evaluations primarily adopt a deductive paradigm, where agents execute tasks based on explicitly provided rules and static goals, often within limited planning horizons. Crucially, this neglects the inductive necessity for agents to discover latent transition laws from experience autonomously, which is the cornerstone for enabling agentic foresight and sustaining strategic coherence. To bridge this gap, we introduce OdysseyArena, which re-centers agent evaluation on long-horizon, active, and inductive interactions. We formalize and instantiate four primitives, translating abstract transition dynamics into concrete interactive environments. Building upon this, we establish OdysseyArena-Lite for standardized benchmarking, providing a set of 120 tasks to measure an agent's inductive efficiency and long-horizon discovery. Pushing further, we introduce OdysseyArena-Challenge to stress-test agent stability across extreme interaction horizons (e.g., > 200 steps). Extensive experiments on 15+ leading LLMs reveal that even frontier models exhibit a deficiency in inductive scenarios, identifying a critical bottleneck in the pursuit of autonomous discovery in complex environments. Project homepage: this https URL
| Comments: | 38 pages. Updated author order and added project homepage |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2602.05843 [cs.CL] |
| (or arXiv:2602.05843v3 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2602.05843 arXiv-issued DOI via DataCite |
Submission history
From: Hang Yan [view email]
[v1]
Thu, 5 Feb 2026 16:31:43 UTC (6,260 KB)
[v2]
Thu, 4 Jun 2026 17:59:52 UTC (4,931 KB)
[v3]
Sat, 24 Oct 2026 10:26:13 UTC (4,940 KB)
来源：arXiv cs.CL · arxiv.org