DeepSearch-World：可驗證環境中深度搜尋代理的自我蒸餾

摘要

訓練工具使用代理從自身經驗中改進仍然具有挑戰性，因為監督式微調依賴於固定的教師蒸餾軌跡，而稀疏獎勵的強化學習則為長程互動提供了薄弱的監督訊號。我們提出DeepSearch-Evolve，這是一個基於DeepSearch-World的網路代理自我蒸餾框架，後者是一個具備可複現搜尋與頁面閱讀工具的確定性可驗證環境。DeepSearch-World包含從實體層級隨機漫步建構的42萬個多跳問答任務，並支援多種對自我演化有用的關鍵代理認知行為，包括進度驗證、落地反思與失敗復原。DeepSearch-Evolve反覆執行軌跡生成、篩選、數據混合與微調，以訓練更強大的代理。無須從更強模型蒸餾，DeepSearch-World-9B在開源代理中達到競爭性表現，在BrowseComp上獲得31.2%，在GAIA上獲得61.5%，在HotpotQA上獲得93.4%，顯示可驗證環境能讓長程網路代理實現可擴展的自我演化。我們將釋出該環境、42萬訓練集、驗證集、模型與程式碼，以促進未來對自我改進深度搜尋代理的研究。

English

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.