S1-DeepResearch: Beyond Search, Toward Real-World Long-Horizon Research Agents

In three linesS1-DeepResearch introduces a unified trajectory construction paradigm for deep research agents combining closed-ended QA and open-ended exploration. The 32B model achieves SOTA performance among open-source models on 20 benchmarks spanning complex reasoning, knowledge synthesis, report generation, and file understanding.

Summary generated by Claude — human-verified