🚀 We are excited to share Agents-A1 from the Shanghai AI Lab.</p>\n<p>Agents-A1 is a 35B MoE agentic model designed to scale long-horizon scientific and engineering capabilities, rather than simply scaling model parameters. It learns from knowledge-action trajectories that connect reasoning, tool use, execution feedback, and verification.</p>\n<p>🔬 Agents-A1 shows strong capabilities in scientific reasoning, research-level coding, ML engineering, and scientific tool use. In our technical report, it achieves competitive results on benchmarks such as HLE with tools, HiPhO, FrontierScience, SciCode, MLE-Bench-Lite, MatTools, and MolBench-Bind.</p>\n<p>🛠️ We hope Agents-A1 can serve as a practical open model for the community to explore autonomous research workflows, tool-integrated scientific problem solving, and next-generation AI-for-Science agents.</p>\n","updatedAt":"2026-06-30T03:24:54.143Z","author":{"_id":"667cf204268f6622dac71961","avatarUrl":"/avatars/90e1928beb2a685e82e19758e4a6b7ae.svg","fullname":"shiyang","name":"sY713","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"it","probability":0.2248600274324417},"editors":["sY713"],"editorAvatarUrls":["/avatars/90e1928beb2a685e82e19758e4a6b7ae.svg"],"reactions":[{"reaction":"🔥","users":["Ryanbihao","sY713","DubbyDu","nybchen","huangst","ChefCurryGoat"],"count":6}],"isReport":false}},{"id":"6a4336e678897084ba9a746a","author":{"_id":"667cf204268f6622dac71961","avatarUrl":"/avatars/90e1928beb2a685e82e19758e4a6b7ae.svg","fullname":"shiyang","name":"sY713","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false},"createdAt":"2026-06-30T03:24:22.000Z","type":"comment","data":{"edited":true,"hidden":true,"hiddenBy":"","latest":{"raw":"This comment has been hidden","html":"This comment has been hidden","updatedAt":"2026-06-30T03:24:48.679Z","author":{"_id":"667cf204268f6622dac71961","avatarUrl":"/avatars/90e1928beb2a685e82e19758e4a6b7ae.svg","fullname":"shiyang","name":"sY713","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"editors":[],"editorAvatarUrls":[],"reactions":[]}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.30616","authors":[{"_id":"6a432d66763f63ca3757e869","name":"Lei Bai","hidden":false},{"_id":"6a432d66763f63ca3757e86a","name":"Zongsheng Cao","hidden":false},{"_id":"6a432d66763f63ca3757e86b","name":"Yang Chen","hidden":false},{"_id":"6a432d66763f63ca3757e86c","name":"Zhiyao Cui","hidden":false},{"_id":"6a432d66763f63ca3757e86d","name":"Shangheng Du","hidden":false},{"_id":"6a432d66763f63ca3757e86e","name":"Yue Fan","hidden":false},{"_id":"6a432d66763f63ca3757e86f","name":"Shiyang Feng","hidden":false},{"_id":"6a432d66763f63ca3757e870","name":"Zijie Guo","hidden":false},{"_id":"6a432d66763f63ca3757e871","name":"Haonan He","hidden":false},{"_id":"6a432d66763f63ca3757e872","name":"Liang He","hidden":false},{"_id":"6a432d66763f63ca3757e873","name":"Xiaohan He","hidden":false},{"_id":"6a432d66763f63ca3757e874","name":"Shuyue Hu","hidden":false},{"_id":"6a432d66763f63ca3757e875","name":"Yusong Hu","hidden":false},{"_id":"6a432d66763f63ca3757e876","name":"Songtao Huang","hidden":false},{"_id":"6a432d66763f63ca3757e877","name":"Yichen Jiang","hidden":false},{"_id":"6a432d66763f63ca3757e878","name":"Hao Li","hidden":false},{"_id":"6a432d66763f63ca3757e879","name":"Xin Li","hidden":false},{"_id":"6a432d66763f63ca3757e87a","name":"Dahua Lin","hidden":false},{"_id":"6a432d66763f63ca3757e87b","name":"Weihao Lin","hidden":false},{"_id":"6a432d66763f63ca3757e87c","name":"Fenghua Ling","hidden":false},{"_id":"6a432d66763f63ca3757e87d","name":"Dongrui Liu","hidden":false},{"_id":"6a432d66763f63ca3757e87e","name":"Zhuo Liu","hidden":false},{"_id":"6a432d66763f63ca3757e87f","name":"Runmin Ma","hidden":false},{"_id":"6a432d66763f63ca3757e880","name":"Chunjiang Mu","hidden":false},{"_id":"6a432d66763f63ca3757e881","name":"Haoyang Peng","hidden":false},{"_id":"6a432d66763f63ca3757e882","name":"Tianshuo Peng","hidden":false},{"_id":"6a432d66763f63ca3757e883","name":"Jinxin Shi","hidden":false},{"_id":"6a432d66763f63ca3757e884","name":"Luohe Shi","hidden":false},{"_id":"6a432d66763f63ca3757e885","name":"Boyuan Sun","hidden":false},{"_id":"6a432d66763f63ca3757e886","name":"Zelin Tan","hidden":false},{"_id":"6a432d66763f63ca3757e887","name":"Shengji Tang","hidden":false},{"_id":"6a432d66763f63ca3757e888","name":"Qianyi Wang","hidden":false},{"_id":"6a432d66763f63ca3757e889","name":"Yiming Wu","hidden":false},{"_id":"6a432d66763f63ca3757e88a","name":"Yi Xie","hidden":false},{"_id":"6a432d66763f63ca3757e88b","name":"Xiangchao Yan","hidden":false},{"_id":"6a432d66763f63ca3757e88c","name":"Jingqi Ye","hidden":false},{"_id":"6a432d66763f63ca3757e88d","name":"Peng Ye","hidden":false},{"_id":"6a432d66763f63ca3757e88e","name":"Fangchen Yu","hidden":false},{"_id":"6a432d66763f63ca3757e88f","name":"Jiakang Yuan","hidden":false},{"_id":"6a432d66763f63ca3757e890","name":"Bihao Zhan","hidden":false},{"_id":"6a432d66763f63ca3757e891","name":"Bo Zhang","hidden":false},{"_id":"6a432d66763f63ca3757e892","name":"Chen Zhang","hidden":false},{"_id":"6a432d66763f63ca3757e893","name":"Shufei Zhang","hidden":false},{"_id":"6a432d66763f63ca3757e894","name":"Shuaiyu Zhang","hidden":false},{"_id":"6a432d66763f63ca3757e895","name":"Wenlong Zhang","hidden":false},{"_id":"6a432d66763f63ca3757e896","name":"Yiqun Zhang","hidden":false},{"_id":"6a432d66763f63ca3757e897","name":"Junpeng Zhao","hidden":false},{"_id":"6a432d66763f63ca3757e898","name":"Zhijie Zhong","hidden":false},{"_id":"6a432d66763f63ca3757e899","name":"Bowen Zhou","hidden":false},{"_id":"6a432d66763f63ca3757e89a","name":"Yuhao Zhou","hidden":false}],"publishedAt":"2026-06-29T00:00:00.000Z","submittedOnDailyAt":"2026-06-30T00:00:00.000Z","title":"Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent","submittedOnDailyBy":{"_id":"667cf204268f6622dac71961","avatarUrl":"/avatars/90e1928beb2a685e82e19758e4a6b7ae.svg","isPro":false,"fullname":"shiyang","user":"sY713","type":"user","name":"sY713"},"summary":"We introduce Agents-A1, a 35B Mixture-of-Experts Agentic Model that reaches trillion-parameter-level performance by scaling the agent horizon. We investigate agent-horizon scaling from two perspectives: scaling long-horizon trajectories and scaling heterogeneous agent abilities. To support this goal, we build a long-horizon knowledge-action infrastructure that connects external knowledge, actions, observations, and verifier outcomes, producing agentic trajectories with an average length of 45K tokens. Based on this, we train Agents-A1 with a three-stage recipe. First, we perform full-domain supervised fine-tuning to align the base model with broad agentic behaviors. Second, we train domain-level teacher models to capture specialized expertise in each domain. Third, we propose a multi-teacher domain-routed on-policy distillation with salient vocabulary alignment to improve knowledge transfer efficiency across different domains, unifying six heterogeneous domains into one deployable student model. Agents-A1 achieves strong and broad performance for long-horizon agent benchmarks. Compared with 1T-parameter model such as Kimi-K2.6 and DeepSeek-V4-pro, Agents-A1 achieves leading results on SEAL-0 (56.4), IFBench (80.6), HiPhO (46.4), FrontierScience-Olympiad (79.0), and MolBench-Bind (56.8), and remains highly competitive on SciCode (44.3), HLE (47.6) and BrowseComp (75.5). We hope this work provides the community with a practical path for scaling the horizon using a 35B agent that can reach or match the performance of 1T models on long-horizon tasks.","upvotes":67,"discussionId":"6a432d66763f63ca3757e89b","projectPage":"https://internscience.github.io/Agents-A1/","githubRepo":"https://github.com/InternScience/Agents-A1","githubRepoAddedBy":"user","ai_summary":"Agents-A1, a 35B Mixture-of-Experts Agentic Model, achieves trillion-parameter-level performance through long-horizon trajectory scaling and heterogeneous agent ability scaling via a three-stage training approach involving supervised fine-tuning, domain-level teacher models, and multi-teacher distillation.","ai_keywords":["Mixture-of-Experts","agentic model","agent horizon","long-horizon trajectories","heterogeneous agent abilities","knowledge-action infrastructure","agentic trajectories","supervised fine-tuning","domain-level teacher models","multi-teacher domain-routed on-policy distillation","salient vocabulary alignment","domain-routed training"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":68,"organization":{"_id":"690af7a885f71496ea396393","name":"InternScience","fullname":"Intern Science","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/65f2a4198404ac0e4c0f175f/XIPU4aCPBogXSrj6NrfLk.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"643dfd235aafbdca3a5792c0","avatarUrl":"/avatars/ce8553cf5936012c692e08054ee27937.svg","isPro":false,"fullname":"Bo Zhang","user":"BoZhang","type":"user"},{"_id":"667cf204268f6622dac71961","avatarUrl":"/avatars/90e1928beb2a685e82e19758e4a6b7ae.svg","isPro":false,"fullname":"shiyang","user":"sY713","type":"user"},{"_id":"65b88b92e0bde92c176a888a","avatarUrl":"/avatars/fc1cb54328ca93860e97fc73a3c1eb2f.svg","isPro":false,"fullname":"Xiangchao Yan","user":"yxc97","type":"user"},{"_id":"64e46682468601277e3ab46c","avatarUrl":"/avatars/8fec6340b2543489dba6066e27a6bd6d.svg","isPro":false,"fullname":"Yue Fan","user":"SaviorYue","type":"user"},{"_id":"65d6ac995e971572dafe2576","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65d6ac995e971572dafe2576/4lDwi-j2M9V-myJF_PKyz.png","isPro":false,"fullname":"Shuaiyu Zhang","user":"ChefCurryGoat","type":"user"},{"_id":"6a2da6c8ca070ee12c6e396c","avatarUrl":"/avatars/0355287dcabaa67dbc7f0b10b87451f9.svg","isPro":false,"fullname":"Joe Mama","user":"JoeMama123123123","type":"user"},{"_id":"6538dd471ad9b3ba7c2df861","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6538dd471ad9b3ba7c2df861/MbEa7KHAK6u7PRb7WiPUC.jpeg","isPro":false,"fullname":"Tianshuo Peng","user":"Potentialts","type":"user"},{"_id":"63bab9c1bb6a2fabd14421bd","avatarUrl":"/avatars/c47030cf167072bf6ce3421f025c7746.svg","isPro":false,"fullname":"Yuhao Zhou","user":"Soptq","type":"user"},{"_id":"6712510a164bbd72c709c428","avatarUrl":"/avatars/63e44aa1e9f7f6b1f3a53455f53e85a1.svg","isPro":false,"fullname":"Lingshuai Lin","user":"Lynn031108","type":"user"},{"_id":"677f65919300252ee4093a2c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/4Jix1E_DQHOE8W97lceYU.png","isPro":false,"fullname":"hxh","user":"sunnyhxh","type":"user"},{"_id":"68ad9cb3bcaa8d84217a8bdf","avatarUrl":"/avatars/dbb3199cf5bfc2acdbd38069c823c027.svg","isPro":false,"fullname":"Fangchen Yu","user":"SciYu","type":"user"},{"_id":"67580a78f6a2918c7abc8fb2","avatarUrl":"/avatars/415924307877ba77d666e78ce215772e.svg","isPro":false,"fullname":"Runmin Ma","user":"mrm0107","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":3,"organization":{"_id":"690af7a885f71496ea396393","name":"InternScience","fullname":"Intern Science","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/65f2a4198404ac0e4c0f175f/XIPU4aCPBogXSrj6NrfLk.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.30616.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2606.30616 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2606.30616 in a Space README.md to link it from this page.