complex-reasoning

This paper introduces S1-DeepResearch-32B, an open-source model and 15K trajectory dataset for deep research agents, achieving state-of-the-art performance across 20 benchmarks by jointly modeling information acquisition, knowledge synthesis, and planning.