Reproduce “Axion 1.0: Open Agent Foundation Models, Simplified” 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
Yuqi Jia, Yunjie Tian, Jiayu Zheng, Yijun Liang, Jingchen Sun, Mengyu Zhao, Di Fu
Axion 1.0 trains open agents through two supervised fine-tuning stages, followed by reinforcement learning, without dedicated mid-training, direct preference optimization, or on-policy distillation. It constructs verified tasks for self-contained reasoning and tool-dependent investigation, and releases its data and training process. Axion 1.0-35B scores 86.0% on FrontierScience Olympiad versus Agents-A1’s 79.0%, an indicative comparison, not a controlled head-to-head test.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.