Reproduce “Scaling Agentic Data for Long-Horizon Terminal Intelligence” 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
Shun Zou, Ziyu Ma, Yi Zou, Yong Wang, Lin Chen, Zehui Chen, Guanhua Chen, Xiangxiang Chu and 1 more
TerminalHorizon synthesizes long-horizon terminal-agent training tasks by grounding task seeds in professional evidence, challenging constructed tasks with an independent agent, repairing shortcuts, and expanding accepted tasks with requirements linked to prior results. Fine-tuning Qwen3.5-27B on 2,852 trajectories reaches 55.8% on Terminal-Bench 2.0, versus 41.6% for its base model; this does not isolate either synthesis stage.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.