https://github.com/kyegomez/awesome-multi-agent-papers

\n","updatedAt":"2025-08-20T15:35:52.138Z","author":{"_id":"6270324ebecab9e2dcf245de","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6270324ebecab9e2dcf245de/cMbtWSasyNlYc9hvsEEzt.jpeg","fullname":"Kye Gomez","name":"kye","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":33}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8897700309753418},"editors":["kye"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/6270324ebecab9e2dcf245de/cMbtWSasyNlYc9hvsEEzt.jpeg"],"reactions":[],"isReport":false}},{"id":"68a6782854b379838b94c9f0","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":264},"createdAt":"2025-08-21T01:36:40.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"This is an automated message from the [Librarian Bot](https://huggingface.co/librarian-bots). I found the following papers similar to this paper. \n\nThe following papers were recommended by the Semantic Scholar API \n\n* [LLM Collaboration With Multi-Agent Reinforcement Learning](https://huggingface.co/papers/2508.04652) (2025)\n* [Reducing Cognitive Load in Multi-Agent Reinforcement Learning for Mathematical Problem Solving: Decoupling Reasoning and Code Generation](https://huggingface.co/papers/2508.08882) (2025)\n* [EvoAgentX: An Automated Framework for Evolving Agentic Workflows](https://huggingface.co/papers/2507.03616) (2025)\n* [GoalfyMax: A Protocol-Driven Multi-Agent System for Intelligent Experience Entities](https://huggingface.co/papers/2507.09497) (2025)\n* [Agent KB: Leveraging Cross-Domain Experience for Agentic Problem Solving](https://huggingface.co/papers/2507.06229) (2025)\n* [Deep Research Agents: A Systematic Examination And Roadmap](https://huggingface.co/papers/2506.18096) (2025)\n\n\n Please give a thumbs up to this comment if you found it helpful!\n\n If you want recommendations for any Paper on Hugging Face checkout [this](https://huggingface.co/spaces/librarian-bots/recommend_similar_papers) Space\n\n You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: `@librarian-bot recommend`","html":"

This is an automated message from the Librarian Bot. I found the following papers similar to this paper.

The following papers were recommended by the Semantic Scholar API

Please give a thumbs up to this comment if you found it helpful!

If you want recommendations for any Paper on Hugging Face checkout this Space

You can directly ask Librarian Bot for paper recommendations by tagging it in a comment: \n\n@librarian-bot\n\t recommend

\n","updatedAt":"2025-08-21T01:36:40.485Z","author":{"_id":"63d3e0e8ff1384ce6c5dd17d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg","fullname":"Librarian Bot (Bot)","name":"librarian-bot","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":264}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7296595573425293},"editors":["librarian-bot"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674830754237-63d3e0e8ff1384ce6c5dd17d.jpeg"],"reactions":[{"reaction":"👍","users":["tianyue818"],"count":1}],"isReport":false}},{"id":"68aa029b0f722221d1b643ca","author":{"_id":"65d9fc2a0e6ad24551d87a1e","avatarUrl":"/avatars/3aedb9522cc3cd08349d654f523fd792.svg","fullname":"Grant Singleton","name":"grantsing","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1},"createdAt":"2025-08-23T18:04:11.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"arXiv explained breakdown of this paper 👉 https://arxivexplained.com/papers/chain-of-agents-end-to-end-agent-foundation-models-via-multi-agent-distillation-and-agentic-rl","html":"

arXiv explained breakdown of this paper 👉 https://arxivexplained.com/papers/chain-of-agents-end-to-end-agent-foundation-models-via-multi-agent-distillation-and-agentic-rl

\n","updatedAt":"2025-08-23T18:04:11.168Z","author":{"_id":"65d9fc2a0e6ad24551d87a1e","avatarUrl":"/avatars/3aedb9522cc3cd08349d654f523fd792.svg","fullname":"Grant Singleton","name":"grantsing","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7062282562255859},"editors":["grantsing"],"editorAvatarUrls":["/avatars/3aedb9522cc3cd08349d654f523fd792.svg"],"reactions":[],"isReport":false}},{"id":"68ad48e02da83956bd68088c","author":{"_id":"679d3a280e751b0476f50b48","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/FMF4sJxbxvQYIabfpD8qO.jpeg","fullname":"Sharon Agwu","name":"shagwu22","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false},"createdAt":"2025-08-26T05:40:48.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"👍this is fire ","html":"

👍this is fire

\n","updatedAt":"2025-08-26T05:40:48.109Z","author":{"_id":"679d3a280e751b0476f50b48","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/FMF4sJxbxvQYIabfpD8qO.jpeg","fullname":"Sharon Agwu","name":"shagwu22","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9528868198394775},"editors":["shagwu22"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/FMF4sJxbxvQYIabfpD8qO.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2508.13167","authors":[{"_id":"68a535f16cf0bf898542ec6c","name":"Weizhen Li","hidden":false},{"_id":"68a535f16cf0bf898542ec6d","name":"Jianbo Lin","hidden":false},{"_id":"68a535f16cf0bf898542ec6e","name":"Zhuosong Jiang","hidden":false},{"_id":"68a535f16cf0bf898542ec6f","name":"Jingyi Cao","hidden":false},{"_id":"68a535f16cf0bf898542ec70","name":"Xinpeng Liu","hidden":false},{"_id":"68a535f16cf0bf898542ec71","name":"Jiayu Zhang","hidden":false},{"_id":"68a535f16cf0bf898542ec72","name":"Zhenqiang Huang","hidden":false},{"_id":"68a535f16cf0bf898542ec73","name":"Qianben Chen","hidden":false},{"_id":"68a535f16cf0bf898542ec74","name":"Weichen Sun","hidden":false},{"_id":"68a535f16cf0bf898542ec75","name":"Qiexiang Wang","hidden":false},{"_id":"68a535f16cf0bf898542ec76","name":"Hongxuan Lu","hidden":false},{"_id":"68a535f16cf0bf898542ec77","user":{"_id":"64301abe450c0de9a1d3d18e","avatarUrl":"/avatars/01b284874dadc7d21d656c53dcb77e42.svg","isPro":false,"fullname":"tianrui","user":"tianyue818","type":"user"},"name":"Tianrui Qin","status":"claimed_verified","statusLastChangedAt":"2025-08-20T08:51:31.629Z","hidden":false},{"_id":"68a535f16cf0bf898542ec78","name":"Chenghao Zhu","hidden":false},{"_id":"68a535f16cf0bf898542ec79","user":{"_id":"65d17ddfbf849f3b796e49ad","avatarUrl":"/avatars/d9d89666cf9f44fe6b88118b8be50ed0.svg","isPro":false,"fullname":"yycsu","user":"yycsu","type":"user"},"name":"Yi Yao","status":"claimed_verified","statusLastChangedAt":"2025-09-26T12:24:17.710Z","hidden":false},{"_id":"68a535f16cf0bf898542ec7a","name":"Shuying Fan","hidden":false},{"_id":"68a535f16cf0bf898542ec7b","user":{"_id":"68022a47806e99452af4d05e","avatarUrl":"/avatars/4546956c787242b9950a05cd6ad26a21.svg","isPro":false,"fullname":"wanwan","user":"wanwan1212","type":"user"},"name":"Xiaowan Li","status":"claimed_verified","statusLastChangedAt":"2025-08-20T08:51:19.190Z","hidden":false},{"_id":"68a535f16cf0bf898542ec7c","name":"Tiannan Wang","hidden":false},{"_id":"68a535f16cf0bf898542ec7d","name":"Pai Liu","hidden":false},{"_id":"68a535f16cf0bf898542ec7e","user":{"_id":"6578265ddea7e2122d02f6ba","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6578265ddea7e2122d02f6ba/Bh6JjoVF5ceLSjV7Z7nTk.jpeg","isPro":false,"fullname":"kang zhu","user":"kangz","type":"user"},"name":"King Zhu","status":"claimed_verified","statusLastChangedAt":"2025-08-20T08:51:21.203Z","hidden":false},{"_id":"68a535f16cf0bf898542ec7f","name":"He Zhu","hidden":false},{"_id":"68a535f16cf0bf898542ec80","user":{"_id":"657c1f7e688f1a0f7ecfe264","avatarUrl":"/avatars/265afcb7b0eeddbcf66ec4cdd4920dd3.svg","isPro":false,"fullname":"Dingfeng Shi","user":"hugteste","type":"user"},"name":"Dingfeng Shi","status":"claimed_verified","statusLastChangedAt":"2025-08-20T08:51:24.492Z","hidden":false},{"_id":"68a535f16cf0bf898542ec81","name":"Piaohong Wang","hidden":false},{"_id":"68a535f16cf0bf898542ec82","name":"Yeyi Guan","hidden":false},{"_id":"68a535f16cf0bf898542ec83","name":"Xiangru Tang","hidden":false},{"_id":"68a535f16cf0bf898542ec84","user":{"_id":"6417d9ea8f689506e7148417","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6417d9ea8f689506e7148417/bAYcruWNw4WvmuQcGgcwC.jpeg","isPro":false,"fullname":"minghao","user":"Liam-Liu","type":"user"},"name":"Minghao Liu","status":"claimed_verified","statusLastChangedAt":"2025-08-20T08:51:27.711Z","hidden":false},{"_id":"68a535f16cf0bf898542ec85","name":"Yuchen Eleanor Jiang","hidden":false},{"_id":"68a535f16cf0bf898542ec86","name":"Jian Yang","hidden":false},{"_id":"68a535f16cf0bf898542ec87","name":"Jiaheng Liu","hidden":false},{"_id":"68a535f16cf0bf898542ec88","user":{"_id":"638efcf4c67af472d316d424","avatarUrl":"/avatars/97a57859d7d87a3a8f1bb41d32a72bc2.svg","isPro":false,"fullname":"Ge Zhang","user":"zhangysk","type":"user"},"name":"Ge Zhang","status":"claimed_verified","statusLastChangedAt":"2025-08-20T08:51:36.844Z","hidden":false},{"_id":"68a535f16cf0bf898542ec89","name":"Wangchunshu Zhou","hidden":false}],"publishedAt":"2025-08-06T17:01:02.000Z","submittedOnDailyAt":"2025-08-20T01:12:39.713Z","title":"Chain-of-Agents: End-to-End Agent Foundation Models via Multi-Agent\n Distillation and Agentic RL","submittedOnDailyBy":{"_id":"628c8598ef14f971b698107f","avatarUrl":"/avatars/3a4ad87e6b5f9e836a1160d869df1447.svg","isPro":false,"fullname":"Zhou","user":"Wangchunshu","type":"user"},"summary":"Recent advances in large language models (LLMs) and multi-agent systems have\ndemonstrated remarkable capabilities in complex problem-solving tasks such as\ndeep research, vibe coding, and mathematical reasoning. However, most existing\nmulti-agent systems are built upon manual prompt/workflow engineering with\nsophisticated agent frameworks, making them computationally inefficient, less\ncapable, and can not benefit from data-centric learning. In this work, we\nintroduce Chain-of-Agents (CoA), a novel paradigm of LLM reasoning that enables\nnative end-to-end complex problem-solving in the same way as a multi-agent\nsystem (i.e., multi-turn problem solving with multiple tools and multiple\nagents) within one model. In chain-of-agents problem-solving, the model\ndynamically activates different tool agents and role-playing agents to simulate\nmulti-agent collaboration in an end-to-end fashion. To elicit end-to-end\nchain-of-agents problem-solving abilities in LLMs, we introduce a multi-agent\ndistillation framework to distill state-of-the-art multi-agent systems into\nchain-of-agents trajectories for agentic supervised fine-tuning. We then use\nagentic reinforcement learning on verifiable agentic tasks to further improve\nthe models' capabilities on chain-of-agents problem solving. We call the\nresulting models Agent Foundation Models (AFMs). Our empirical studies\ndemonstrate that AFM establishes new state-of-the-art performance across\ndiverse benchmarks in both web agent and code agent settings. We make the\nentire research, including the model weights, code for training and evaluation,\nand the training data, fully open-sourced, which offers a solid starting point\nfor future research on agent models and agentic RL.","upvotes":127,"discussionId":"68a535f16cf0bf898542ec8a","projectPage":"https://chain-of-agents-afm.github.io/","githubRepo":"https://github.com/OPPO-PersonalAI/Agent_Foundation_Models","ai_summary":"Chain-of-Agents (CoA) paradigm enables end-to-end complex problem-solving in LLMs through dynamic agent activation, improving performance via multi-agent distillation and agentic reinforcement learning.","ai_keywords":["large language models","multi-agent systems","deep research","vibe coding","mathematical reasoning","prompt/workflow engineering","agent frameworks","chain-of-agents","tool agents","role-playing agents","multi-agent distillation","agentic supervised fine-tuning","agentic reinforcement learning","Agent Foundation Models","AFMs","web agent","code agent","verifiable agentic tasks"],"githubStars":432},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"628c8598ef14f971b698107f","avatarUrl":"/avatars/3a4ad87e6b5f9e836a1160d869df1447.svg","isPro":false,"fullname":"Zhou","user":"Wangchunshu","type":"user"},{"_id":"632bfaebea6e62428ab0e9c2","avatarUrl":"/avatars/344aaf371bbba9aea091b12741c451e5.svg","isPro":false,"fullname":"Tiannan Wang","user":"WTNswaggy","type":"user"},{"_id":"654691677e0f913218108438","avatarUrl":"/avatars/b1792d5fca7ab338d8c64bc616ac45ec.svg","isPro":false,"fullname":"Pai","user":"Pai3dot14","type":"user"},{"_id":"64301abe450c0de9a1d3d18e","avatarUrl":"/avatars/01b284874dadc7d21d656c53dcb77e42.svg","isPro":false,"fullname":"tianrui","user":"tianyue818","type":"user"},{"_id":"64be3a8b86e7fb5b8a7ec8a9","avatarUrl":"/avatars/58280c82f643a5a8073623eff33fefb2.svg","isPro":false,"fullname":"Chen","user":"Qianben","type":"user"},{"_id":"6773a30da3933b9326bb269e","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/7BoR9p_kQB89fRqGetll7.png","isPro":false,"fullname":"xinpeng","user":"liuxinpeng353","type":"user"},{"_id":"66a1a936e3d75c887db54ff2","avatarUrl":"/avatars/bfe10fcfed301335a5c3f8e8b80ee229.svg","isPro":false,"fullname":"Wong","user":"AllenGZ","type":"user"},{"_id":"6850e5f7e4b48f282cf35d69","avatarUrl":"/avatars/7cef6afa8f6aec3ced7269d986c6e91d.svg","isPro":false,"fullname":"Sun Weichen","user":"ChrisHs","type":"user"},{"_id":"67f91c613e25f1cc591082d6","avatarUrl":"/avatars/1923d6dc3297966f23a82a684723149b.svg","isPro":false,"fullname":"Weizhen Li","user":"weizhenli","type":"user"},{"_id":"66a4f596d8e85b03def67c33","avatarUrl":"/avatars/7029923a019b2e972a157e1276a8089c.svg","isPro":false,"fullname":"Xinxin Liu","user":"AsahiRokkaLOCK","type":"user"},{"_id":"66d6bac1b61dd110228d90e7","avatarUrl":"/avatars/409f1c65c161a1adc76e4ae3c732be45.svg","isPro":false,"fullname":"huang","user":"gumbou","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"}],"acceptLanguages":["*"],"dailyPaperRank":1}">

Papers

arxiv:2508.13167

Chain-of-Agents: End-to-End Agent Foundation Models via Multi-Agent Distillation and Agentic RL

Published on Aug 6

· Submitted by

Zhou on Aug 20

#1 Paper of the day

Upvote

127

Authors:

Tianrui Qin ,

Yi Yao ,

Xiaowan Li ,

King Zhu ,

Dingfeng Shi ,

Abstract

Chain-of-Agents (CoA) paradigm enables end-to-end complex problem-solving in LLMs through dynamic agent activation, improving performance via multi-agent distillation and agentic reinforcement learning.

AI-generated summary

Recent advances in large language models (LLMs) and multi-agent systems have demonstrated remarkable capabilities in complex problem-solving tasks such as deep research, vibe coding, and mathematical reasoning. However, most existing multi-agent systems are built upon manual prompt/workflow engineering with sophisticated agent frameworks, making them computationally inefficient, less capable, and can not benefit from data-centric learning. In this work, we introduce Chain-of-Agents (CoA), a novel paradigm of LLM reasoning that enables native end-to-end complex problem-solving in the same way as a multi-agent system (i.e., multi-turn problem solving with multiple tools and multiple agents) within one model. In chain-of-agents problem-solving, the model dynamically activates different tool agents and role-playing agents to simulate multi-agent collaboration in an end-to-end fashion. To elicit end-to-end chain-of-agents problem-solving abilities in LLMs, we introduce a multi-agent distillation framework to distill state-of-the-art multi-agent systems into chain-of-agents trajectories for agentic supervised fine-tuning. We then use agentic reinforcement learning on verifiable agentic tasks to further improve the models' capabilities on chain-of-agents problem solving. We call the resulting models Agent Foundation Models (AFMs). Our empirical studies demonstrate that AFM establishes new state-of-the-art performance across diverse benchmarks in both web agent and code agent settings. We make the entire research, including the model weights, code for training and evaluation, and the training data, fully open-sourced, which offers a solid starting point for future research on agent models and agentic RL.