\n<li>The model weights have been publicly released at: <a href=\"https://huggingface.co/collections/BAAI/arex\">https://huggingface.co/collections/BAAI/arex</a></li>\n<li>The models have also been integrated into our online application: <a href=\"https://arex-research.com/\" rel=\"nofollow\">https://arex-research.com/</a></li>\n</ul>\n","updatedAt":"2026-07-24T02:58:54.364Z","author":{"_id":"64a38c590111d5ff6c3d5f2b","avatarUrl":"/avatars/ef13dc7ce243819bc0da9b04e778b432.svg","fullname":"zhengliu","name":"lz1001","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":15,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9362091422080994},"editors":["lz1001"],"editorAvatarUrls":["/avatars/ef13dc7ce243819bc0da9b04e778b432.svg"],"reactions":[{"reaction":"👍","users":["AlphaBIO","lsj2408"],"count":2}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2607.21461","authors":[{"_id":"6a62d3662ee212ed0e2a148f","name":"Shuqi Lu","hidden":false},{"_id":"6a62d3662ee212ed0e2a1490","user":{"_id":"65ccbcfde75fc480714d1d04","avatarUrl":"/avatars/2def06230512892928537cf2d8499563.svg","isPro":false,"fullname":"Chaofan Li","user":"cfli","type":"user","name":"cfli"},"name":"Chaofan Li","status":"claimed_verified","statusLastChangedAt":"2026-07-24T16:45:04.139Z","hidden":false},{"_id":"6a62d3662ee212ed0e2a1491","user":{"_id":"627933c317c069c72a136b35","avatarUrl":"/avatars/f7aef400e47ac84897e6298c7c881c7a.svg","isPro":false,"fullname":"Kun Luo","user":"Lk123","type":"user","name":"Lk123"},"name":"Kun Luo","status":"claimed_verified","statusLastChangedAt":"2026-07-24T08:45:04.312Z","hidden":false},{"_id":"6a62d3662ee212ed0e2a1492","name":"Zhang Zhang","hidden":false},{"_id":"6a62d3662ee212ed0e2a1493","name":"Hui Wang","hidden":false},{"_id":"6a62d3662ee212ed0e2a1494","name":"Hongwang Xiao","hidden":false},{"_id":"6a62d3662ee212ed0e2a1495","name":"Zheng Liu","hidden":false},{"_id":"6a62d3662ee212ed0e2a1496","name":"Lei Xiong","hidden":false},{"_id":"6a62d3662ee212ed0e2a1497","name":"Jiahao Wang","hidden":false},{"_id":"6a62d3662ee212ed0e2a1498","name":"Sen Wang","hidden":false},{"_id":"6a62d3662ee212ed0e2a1499","name":"Xiyan Jiang","hidden":false},{"_id":"6a62d3662ee212ed0e2a149a","user":{"_id":"6777886b8c91ac2d4b705eaf","avatarUrl":"/avatars/ac27c6728d40edd724ce99cb8369808d.svg","isPro":false,"fullname":"Wanli Li","user":"wanlilll","type":"user","name":"wanlilll"},"name":"Wanli Li","status":"claimed_verified","statusLastChangedAt":"2026-07-24T08:45:04.307Z","hidden":false},{"_id":"6a62d3662ee212ed0e2a149b","name":"Yuyang Hu","hidden":false},{"_id":"6a62d3662ee212ed0e2a149c","name":"Hongjin Qian","hidden":false},{"_id":"6a62d3662ee212ed0e2a149d","name":"Bingyu Yan","hidden":false},{"_id":"6a62d3662ee212ed0e2a149e","name":"Ziyi Xia","hidden":false},{"_id":"6a62d3662ee212ed0e2a149f","name":"Yingxia Shao","hidden":false},{"_id":"6a62d3662ee212ed0e2a14a0","name":"Kang Liu","hidden":false},{"_id":"6a62d3662ee212ed0e2a14a1","name":"Zhicheng Dou","hidden":false},{"_id":"6a62d3662ee212ed0e2a14a2","name":"Di He","hidden":false},{"_id":"6a62d3662ee212ed0e2a14a3","name":"Chaozhuo Li","hidden":false},{"_id":"6a62d3662ee212ed0e2a14a4","name":"Qiwei Ye","hidden":false},{"_id":"6a62d3662ee212ed0e2a14a5","name":"Zhongyuan Wang","hidden":false},{"_id":"6a62d3662ee212ed0e2a14a6","name":"Zheng Liu","hidden":false}],"publishedAt":"2026-07-23T00:00:00.000Z","submittedOnDailyAt":"2026-07-24T00:00:00.000Z","title":"AREX: Towards a Recursively Self-Improving Agent for Deep Research","submittedOnDailyBy":{"_id":"64a38c590111d5ff6c3d5f2b","avatarUrl":"/avatars/ef13dc7ce243819bc0da9b04e778b432.svg","isPro":false,"fullname":"zhengliu","user":"lz1001","type":"user","name":"lz1001"},"summary":"Deep research requires agents to find answers that jointly satisfy multiple constraints. Discovering such answers is costly, whereas verifying a candidate can often be decomposed into tractable constraint-wise checks. This discovery--verification asymmetry suggests that a research agent should do more than simply search longer: it should recursively improve its current answer by verifying intermediate results and using the partially verified state to guide subsequent refinement. We introduce AREX, a family of Recursively Self-Improving (RSI) deep research agents. AREX alternates between an inner research loop that gathers evidence and constructs a provisional answer, and an outer self-improvement loop that audits the answer constraint-wise, identifies unresolved claims, and launches targeted follow-up research. To sustain RSI over long horizons, AREX learns an autonomous context-update tool that compresses growing interaction history into a compact improvement state preserving verified evidence and unresolved constraints, without relying on an external model. We train AREX on verified synthetic tasks and high-quality trajectories through agentic mid-training and long-horizon reinforcement learning. To mitigate sparse final rewards during long horizon learning, we emphasize key steps where decisive evidence is acquired or erroneous research directions are corrected. We instantiate a dense 4B model and a 122B-A10B Mixture-of-Experts model. Across BrowseComp, WideSearch, DeepSearchQA, Humanity's Last Exam (HLE), and other reasoning and tool-use benchmarks, AREX substantially outperforms comparable-scale baselines and remains competitive with models using substantially more activated parameters.","upvotes":115,"discussionId":"6a62d3672ee212ed0e2a14a7","projectPage":"https://vectorspacelab.github.io/arex-model/","githubRepo":"https://github.com/VectorSpaceLab/arex-model","githubRepoAddedBy":"user","githubStars":11,"organization":{"_id":"61be9739d2f9358e24ca0a4f","name":"BAAI","fullname":"Beijing Academy of Artificial Intelligence","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1664511063789-632c234f42c386ebd2710434.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64a38c590111d5ff6c3d5f2b","avatarUrl":"/avatars/ef13dc7ce243819bc0da9b04e778b432.svg","isPro":false,"fullname":"zhengliu","user":"lz1001","type":"user"},{"_id":"65ccbcfde75fc480714d1d04","avatarUrl":"/avatars/2def06230512892928537cf2d8499563.svg","isPro":false,"fullname":"Chaofan Li","user":"cfli","type":"user"},{"_id":"627933c317c069c72a136b35","avatarUrl":"/avatars/f7aef400e47ac84897e6298c7c881c7a.svg","isPro":false,"fullname":"Kun Luo","user":"Lk123","type":"user"},{"_id":"665ebae8bcbb98f60db0b4b1","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/665ebae8bcbb98f60db0b4b1/YTKM4qTZXh_2SeU8U7BfB.webp","isPro":false,"fullname":"Jiale Zhao","user":"Heisenburger2000","type":"user"},{"_id":"63211eee8c0da827c72cb783","avatarUrl":"/avatars/07959a3649e5f79c70a054059308f350.svg","isPro":false,"fullname":"Guojiang Zhao","user":"guojianz","type":"user"},{"_id":"64a627232944e255ef574dda","avatarUrl":"/avatars/4c2fd5bf922013fe691c6a3e3fa138a2.svg","isPro":false,"fullname":"Hongjin Qian","user":"TommyChien","type":"user"},{"_id":"63e11aea14846428ce230f09","avatarUrl":"/avatars/d56be72b18a0e34fd1cce74e70013d4b.svg","isPro":false,"fullname":"sabulin","user":"sabulin","type":"user"},{"_id":"6544b9b646dbdeca34ee5f52","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6544b9b646dbdeca34ee5f52/nRx6m1C4wfZ_xSWoBUNJf.png","isPro":false,"fullname":"Yuyang Hu","user":"namespace-ERI","type":"user"},{"_id":"6564a2ceedae9c33b7654a1f","avatarUrl":"/avatars/42f09356a1282896573ccb44830cd327.svg","isPro":false,"fullname":"JUNJIE ZHOU","user":"JUNJIE99","type":"user"},{"_id":"63f4723e21eb234ab7397bb9","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63f4723e21eb234ab7397bb9/YR8unEn1b0bacK6J6gLGf.png","isPro":false,"fullname":"Lei Xiong","user":"Youxll","type":"user"},{"_id":"63f06116f1a47aaea5bd497b","avatarUrl":"/avatars/7d99ffa59c4579599e852a0ffb261268.svg","isPro":false,"fullname":"Guoxin Chen","user":"GuoxinChen","type":"user"},{"_id":"641bd1737c21ab946bf69aff","avatarUrl":"/avatars/83759075ad893a69a0c2cf5493d7e988.svg","isPro":false,"fullname":"xin luo","user":"sienna223","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":1,"organization":{"_id":"61be9739d2f9358e24ca0a4f","name":"BAAI","fullname":"Beijing Academy of Artificial Intelligence","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1664511063789-632c234f42c386ebd2710434.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2607/2607.21461.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2607.21461 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2607.21461 in a Space README.md to link it from this page.