[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84276-en":3,"doc-seo-84276-105":29,"detail-sidebar-cat-0-en-105":83},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},84276,1374391974564,"Clementine","https://ap-avatar.wpscdn.com/avatar/14000253aa45c000a9e?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779874745381141002",8,"Research & Report","DeepSearch-World Self-Distillation for Deep Search Agents in a Verifiable Environment","Training tool-use agents to improve from their own experience remains difficult because supervised fine-tuning depends on fixed distilled teacher trajectories, while sparse-reward reinforcement learning gives weak supervision for long-horizon interactions. DeepSearch-Evolve presents a self-distillation framework for web agents built on DeepSearch-World, a deterministic, verifiable environment with reproducible search and page-reading tools. It provides 420K multi-hop QA tasks and supports progress verification, grounded reflection, and failure recovery to iteratively train stronger agents. ","DeepSearch-World: Self-Distillation for Deep Search Agents in a  \nVerifiable Environment  \nXinyu Geng1,2,* , Xuanhua He1,* , Sixiang Chen2,3,* , Yanjing Xiao1 , Fan Zhang2 , Shijue Huang1 , Haitao Mi2 , Zhenwen Liang2,†, Tianqing Fang2,†, Yi R. Fung1,†  \n1HKUST, 2Tencent, 3HKUST(GZ)  \nCode | Data | Env  \n*Equal contribution †Corresponding authors  \narXiv :2607 .07820v2 [ cs .CL] 13 Jul 2026  \nAbstract  \nTraining tool-use agents to improve from their own experience remains challenging, as supervised fine-tuning relies on fixed teacherdistilled 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 DeepSearchWorld, a deterministic and verifiable environment with reproducible search and pagereading 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.  \nWithout 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.  \n1 Introduction  \nRecent advances in large language models (LLMs) have enabled agents that move beyond passive text generation to plan multi-step tasks, formulate search queries, read documents, browse the web, and refine answers through tool use and iterative reasoning (Schick et al., 2024 ; OpenAI, 2025a ; Google, 2024) . However, enabling these agents to improve autonomously from their own interactions remains a key challenge toward scalable selfevolving agents (He et al., 2025) .  \nA common recipe is supervised fine-tuning (SFT) on positive trajectories distilled from the backbone model itself (Zelikman et al., 2022 ; Zeng et al., 2024) . However, the performance easily saturates after a few optimization steps, bounded by the inherent capability of the backbone model itself and the diversity of the selfcollected trajectories (Song et al., 2025b) . Another way to instantiate self-evolving is to optimize tool-use agents on their own rollouts with verifiable rewards, typically using RL-style objectives such as GRPO (Guo et al., 2025a ; Team et al., 2025a,c) . However, these rewards are typically sparse and trajectory-level, offering little guidance on whether failures arise from query formulation, tool selection, evidence extraction, or answer synthesis (Liu et al., 2025c ; Li et al., 2026) .  \nOn-policy self-distillation (OPSD) alleviates reward sparsity by matching the student’s own rollouts to dense token-level distributions from a finegrained teacher policy (Hübotter et al., 2026 ; Ye et al., 2026 ; Shen et al., 2026) . However, in agentic tool use, such fine-grained supervision lies in actions, including tool selection, evidence verification, search query reformulation, and progress tracking (Gritta et al., 2026 ; Liu et al., 2025b), which requires a deterministic and verifiable interaction environment. Otherwise, the teacher distribution at each tool-use step can be noisy, limiting the direct applicability of OPSD to long-horizon agents.  \nIn summary, as shown in Fig. 1, existing selfevolving pipelines for tool-use agents are lim  \nited by static imitation signals, sparse outcome rewards, or unreliable dense supervision. Longhorizon agents instead require a verifiable environment that can expose process-level supervision over intermediate tool-use decisions. Therefore, we introduce DeepSearc","cbCaimeXOvWa9v0g","https://ap.wps.com/l/cbCaimeXOvWa9v0g","pdf",2821410,1,26,"English","en",105,"# Introduction\n## Motivation and Challenges\n## Proposed Approach\n## Contributions","[{\"question\":\"What makes DeepSearch-Evolve different from relying on distillation from stronger proprietary models?\",\"answer\":\"DeepSearch-Evolve iteratively generates trajectories, uses verified process signals, and absorbs successful behaviors, enabling open-source agents to improve using verified experience rather than synthetic trajectories from more capable external models.\"}]",1784194543,66,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":78,"head_meta":80,"extra_data":82,"updated_unix":27},"deepsearch-world-self-distillation-for-deep-search-agents-in-a-verifiable-environment","",{"@graph":35,"@context":77},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/deepsearch-world-self-distillation-for-deep-search-agents-in-a-verifiable-environment/84276/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71],{"name":72,"@type":73,"acceptedAnswer":74},"What makes DeepSearch-Evolve different from relying on distillation from stronger proprietary models?","Question",{"text":75,"@type":76},"DeepSearch-Evolve iteratively generates trajectories, uses verified process signals, and absorbs successful behaviors, enabling open-source agents to improve using verified experience rather than synthetic trajectories from more capable external models.","Answer","https://schema.org",{"og:url":51,"og:type":79,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":81,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":84},[85,89,93,97,102,107,112,115,120,123,127],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":86,"show_sort_weight":87,"slug":88},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":90,"show_sort_weight":91,"slug":92},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Exam",70,"exam",{"id":98,"doc_module":4,"doc_module_name":45,"category_name":99,"show_sort_weight":100,"slug":101},5,"Comic",60,"comic",{"id":103,"doc_module":4,"doc_module_name":45,"category_name":104,"show_sort_weight":105,"slug":106},6,"Technology",50,"technology",{"id":108,"doc_module":4,"doc_module_name":45,"category_name":109,"show_sort_weight":110,"slug":111},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":113,"slug":114},30,"research-report",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},9,"Religion & Spirituality",20,"religion-spirituality",{"id":118,"doc_module":4,"doc_module_name":45,"category_name":121,"show_sort_weight":118,"slug":122},"World Cup","world-cup",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":124,"slug":126},10,"Lifestyle","lifestyle",{"id":128,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":98,"slug":130},19,"General","general"]