Reproduce “AgentOdyssey: Open-Ended Long-Horizon Text Game Generation for Test-Time Continual Learning Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zheyuan Zhang, Zehao Wen, Alvin Zhang, Andrew Wang, Jianwen Xie, Daniel Khashabi, Tianmin Shu
AgentOdyssey creates long-horizon text games to evaluate agents that learn during interaction with partial observations and changing worlds. It pairs quest progress with diagnostics of memory, knowledge acquisition, exploration, and token use. Across six games, Long Context GPT-5 achieves mean quest completion of 47.7% versus 30.0% for GPT-5 RAG with short-term memory, showing limited progress rather than establishing real-world capability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.