[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86418-en":3,"doc-seo-86418-105":30,"detail-sidebar-cat-0-en-105":92},{"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":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},86418,4398048949847,"Eliana","https://ap-avatar.wpscdn.com/avatar/400002536579ef2da7f?_k=1778318612642679267",8,"Research & Report","CORE-Bench: A Comprehensive Benchmark for Code Retrieval in the Era of Agentic Coding","Code retrieval is becoming central to coding agents, yet agentic coding needs more than mapping a natural-language query to an isolated code snippet. CORE-Bench addresses this mismatch by evaluating retrieval as requirement-driven repository search, covering concrete repository navigation, relevant file/function discovery, and distractor filtering. The benchmark measures three levels: code understanding, issue-to-edit localization, and broader context retrieval. It includes 180K+ queries and 106K+ broader-context relevance labels, with results showing large performance gaps and gains from supervised fine-tuning.","CORE-Bench: A Comprehensive Benchmark for Code Retrieval in the Era  \nof Agentic Coding  \nFuwei Zhang 1 Yanzhao Zhang Mingxin Li Dingkun Long  \nLexiang Hu2 Pengjun Xie Zhao Zhang3 Fuzhen Zhuang 1  \n1Institute of Artificial Intelligence, Beihang University  \n2 State Key Lab of General AI, School of Intelligence Science and Technology, Peking University  \n3 SKLCCSE, School of Computer Science and Engineering, Beihang University, China  \narXiv :2606 . 1 1864v2 [ cs .IR] 13 Jul 2026  \nAbstract  \nCode retrieval is becoming central to coding agents, but agentic coding requires more than matching a natural-language query to an isolated snippet. Given a user request, a coding agent needs to navigate a concrete repository state, locate relevant files and functions, gather supporting context, and filter similar in-repository distractors. Existing code retrieval benchmarks mainly evaluate docstring-to-function or snippet-level matching, thereby missing this requirement-driven repository search problem. To address this gap, we introduce CORE-Bench 1 , a comprehensive benchmark for code retrieval in the era of agentic coding. CORE-Bench evaluates code retrieval ability at three levels: code understanding, issue-to-edit localization, and broader context retrieval. Built from curated code-search tasks and SWE-bench-series instances, COREBench contains over 180K queries and 106K broader-context relevance labels. Experiments with representative embedding models show a sharp drop from traditional code search to code retrieval in agentic coding settings. Simple supervised fine-tuning of existing embedding models significantly improves performance in this setting, suggesting substantial room for further progress.  \n1 Introduction  \nCode retrieval has long served as a bridge between natural language and code. In agentic coding and vibe-coding settings, retrieval becomes part of the coding-agent workflow: given a bug report, feature request, refactoring request, or other user request, an agent must decide which files, functions,  \n1The dataset is available on Hugging Face, and the evaluation code is available on GitHub.  \nand documents are worth inspecting before editing. Current coding-agent harnesses, such as OpenHands (Wang et al., 2025a) and SWE-agent (Yang et al., 2024), usually do not reduce these decisions to a single embedding lookup. Instead, they rely heavily on command-line repository exploration, including ls, grep, find, file reading, and iterative narrowing. These steps show that retrieval remains central in practice, but the retrieval problem shifts toward interactive, requirement-driven repository search. This raises a concrete question: why are retrieval models that perform well on standard code-search benchmarks often less useful for this search process? This mismatch suggests that code retrieval in agentic coding settings may be better understood as requirement-driven repository search, rather than isolated code-snippet matching. Existing benchmarks are not fully aligned with this setting. CodeSearchNet (Husain et al., 2019) and many code tasks in MTEB (Muennighoff et al., 2023) and CoIR (Li et al., 2025) mainly evaluate decontextualized snippet retrieval under fixed corpora and simplified query-document mappings. These benchmarks measure basic natural-language/code alignment but miss several properties that matter for code retrieval in agentic coding: (1) realistic development requests often have a large intent-toimplementation gap; (2) relevant evidence is heterogeneous and scattered across code, configuration files, dependencies, and other repository artifacts;  \n(3) candidate documents are often long functionlevel chunks or documentation snippets, which can dilute the evidence captured by a single embedding; (4) a single request may require multiple edit snippets and related context; and (5) real repositories contain dense local distractors, such as similar wrappers, adapters, and configuration snippets. Strong performance on","cbCaikvHzrIbqutk","https://ap.wps.com/l/cbCaikvHzrIbqutk","pdf",732333,5,1,27,"English","en",105,"# Abstract\n# Introduction\n## CORE-Bench benchmark design\n## Annotation from SWE-bench series\n## Evaluation setup and contributions","[{\"question\":\"What problem does CORE-Bench target in agentic coding?\",\"answer\":\"CORE-Bench targets the gap between traditional code search benchmarks and the requirement-driven repository search needed by coding agents. It focuses on navigating real repository states, locating files/functions, collecting supporting context, and filtering in-repository distractors.\"},{\"question\":\"How many evaluation levels does CORE-Bench include, and what do they test?\",\"answer\":\"CORE-Bench evaluates code retrieval in three levels: code understanding, issue-to-edit localization, and broader context retrieval. The levels range from foundational retrieval to identifying what must be modified and what auxiliary information is needed.\"},{\"question\":\"What dataset scale and labels does CORE-Bench provide?\",\"answer\":\"CORE-Bench contains over 180K queries and about 106K broader-context relevance labels. It is built from curated code-search tasks and SWE-bench-series instances, with repository snapshots aligned to pre-resolution states.\"}]",1784211616,68,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"core-bench-a-comprehensive-benchmark-for-code-retrieval-in-the-era-of-agentic-coding","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/core-bench-a-comprehensive-benchmark-for-code-retrieval-in-the-era-of-agentic-coding/86418/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-07-24","2026-07-16",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"What problem does CORE-Bench target in agentic coding?","Question",{"text":76,"@type":77},"CORE-Bench targets the gap between traditional code search benchmarks and the requirement-driven repository search needed by coding agents. It focuses on navigating real repository states, locating files/functions, collecting supporting context, and filtering in-repository distractors.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How many evaluation levels does CORE-Bench include, and what do they test?",{"text":81,"@type":77},"CORE-Bench evaluates code retrieval in three levels: code understanding, issue-to-edit localization, and broader context retrieval. The levels range from foundational retrieval to identifying what must be modified and what auxiliary information is needed.",{"name":83,"@type":74,"acceptedAnswer":84},"What dataset scale and labels does CORE-Bench provide?",{"text":85,"@type":77},"CORE-Bench contains over 180K queries and about 106K broader-context relevance labels. It is built from curated code-search tasks and SWE-bench-series instances, with repository snapshots aligned to pre-resolution states.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":93},[94,98,102,106,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":20,"slug":138},19,"General","general"]