Paper 2607.07820
DeepSearch-World: Self-Distillation for Deep Search Agents in a Verifiable Environment
- Published
- Jul 2026
- Research lab
- Independent
- Citations
- 1
- GitHub
- 8 stars
01 In brief
Summary
The paper introduces DeepSearch-World, a deterministic and verifiable offline environment for deep search agents, built on a Wikipedia corpus with 420K multi-hop QA tasks.
It also presents DeepSearch-Evolve, a self-distillation framework that iteratively generates, filters, and converts scaffolded trajectories into ReAct-format supervision for student training.
The framework enables agents to improve from their own verified tool-use experience without distillation from stronger models.
DeepSearch-World-9B, trained from Qwen3.5-9B, achieves competitive results on benchmarks like BrowseComp (31.2%), GAIA (61.5%), and HotpotQA (93.4%), significantly outperforming its backbone.
Ablations show the importance of rejection sampling, quality filtering, and scaffold-to-ReAct conversion.
The environment supports process-level verification and grounded reflection, enabling scalable self-evolution for long-horizon web agents.
02 From the paper
Abstract
Training tool-use agents to improve from their own experience remains challenging, as supervised fine-tuning relies on fixed teacher-distilled trajectories, while sparse-reward reinforcement learning provides weak supervision for long-horizon interactions. We present DeepSearch-Evolve, a self-distillation framework for web agents built on DeepSearch-World, a deterministic and verifiable environment with reproducible search and page-reading tools. DeepSearch-World contains 420K multi-hop QA tasks constructed from entity-level random walks and supports key agentic cognitive behaviors useful for self-evolving, including progress verification, grounded reflection, and failure recovery. DeepSearch-Evolve iteratively performs trajectory generation, filtering, data mixing, and fine-tuning to train stronger agents. Without distillation from more capable models, DeepSearch-World-9B achieves competitive performance compared with open-source agents, reaching 31.2% on BrowseComp, 61.5% on GAIA, and 93.4% on HotpotQA, showing that verifiable environments enable scalable self-evolution for long-horizon web agents. We will release the environment, 420K training pool, validation set, model, and code to facilitate future research on self-improving deep search agents.