Reproduce “LongHarness: Advancing Agentic Reasoning over 100M-Token Corpora via Stateful Evidence Management” 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
Songtao Huang, Zhijie Zhong, Xiaohan He, Shuaiyu Zhang, Zhiyin Yu, Li Kang, Heng Zhou, Yutao Fan and 3 more
LongHarness separates source-linked evidence from evolving conclusions, using unresolved gaps to guide searches and revising conclusions when new evidence arrives. On OfficeQA Pro V2, GLM-5.3-Flash with LongHarness scored 61.1% exact match, versus 4.4% for direct reasoning with the same backbone. The direct model received BM25-retrieved passages in a 256K-token window, so this was not the same open-corpus search setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.