Deep research agents aim to tackle complex knowledge-intensive tasks through long-horizon planning, evidence gathering, reasoning, and report generation. Recent progress in search agents has shown strong capabilities in information retrieval and answer verification; however, most existing training datasets remain search-centric, primarily focusing on closed-ended question answering and information localization. Consequently, they mainly train information-seeking behavior and provide limited coverage of key deep research capabilities, such as evidence integration, knowledge synthesis, planning, file understanding, and structured report generation.
To address this, we propose a unified trajectory construction paradigm that combines closed-ended QA and open-ended exploration. The proposed framework consists of graph-grounded task formulation, agentic trajectory rollout, and multi-dimensional trajectory verification, allowing for scalable synthesis of high-quality agentic trajectories encompassing long-chain complex reasoning, deep research instruction following, report writing, file understanding and generation, and skills usage.
Compared to existing search-oriented datasets, our synthesized trajectories place a greater emphasis on knowledge synthesis, complex reasoning, and planning. S1-DeepResearch-32B achieves state-of-the-art performance among open-source models of comparable scale across 20 benchmarks spanning five capability dimensions, including complex reasoning, instruction following, report generation, file understanding, and skills usage. On several challenging deep research benchmarks, it approaches the performance of leading proprietary frontier models. These results highlight the importance of jointly modeling information acquisition, knowledge synthesis, and planning-oriented agent behaviors for building effective deep research agents.
Blogger's Review: This study introduces a novel deep research agent model that significantly enhances reasoning and report generation capabilities by integrating information acquisition with knowledge synthesis. This provides a crucial theoretical foundation for future research and applications, making it worth monitoring its performance in real-world scenarios.