Reproduce “DR-Venus: Towards Frontier Edge-Scale Deep Research Agents with Only 10K Open Data” on OpenResearch

About this paper

Venus Team, Sunhao Dai, Yong Deng, Jinzhen Lin, Yusheng Song, Guoqing Wang, Xiaofeng Wu, Yuqi Zhou and 5 more

The Venus Team at Ant Group developed DR-Venus, a framework for training frontier edge-scale deep research agents using only 10K open data. This approach combines a meticulously filtered supervised fine-tuning stage with an information gain-based reinforcement learning method, enabling a 4B parameter model to achieve competitive performance on deep research benchmarks, sometimes matching or exceeding much larger 30B-class models.

Key insights

• High-quality agentic supervised fine-tuning, especially with turn-aware resampling of long-horizon trajectories, is crucial for establishing a strong foundation and competitive baseline in small deep research agents.
• Dense, turn-level reward signals, such as information gain and format penalties within an IGPO framework, are essential for effective reinforcement learning in long-horizon deep research tasks for small models.
• The latent capability of small deep research agents can be considerably higher than initial evaluations suggest, with test-time scaling (Pass@K) revealing performance that can sometimes surpass larger proprietary models.

Summary by alphaXiv.