Huggingface: <a href=\"https://huggingface.co/collections/TokenRhythm/neohorse-1\">https://huggingface.co/collections/TokenRhythm/neohorse-1</a>; Github: <a href=\"https://github.com/TokenRhythm/NeoHorse\" rel=\"nofollow\">https://github.com/TokenRhythm/NeoHorse</a></p>\n","updatedAt":"2026-09-09T02:39:15.703Z","author":{"_id":"6527c063e86758eb6ca800a1","avatarUrl":"/avatars/9091be87eea518209c1de9eebfa663c0.svg","fullname":"JarvisPei","name":"JarvisPei","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6209907531738281},"editors":["JarvisPei"],"editorAvatarUrls":["/avatars/9091be87eea518209c1de9eebfa663c0.svg"],"reactions":[{"reaction":"🚀","users":["TokenRhythmAdmin"],"count":1}],"isReport":false}},{"id":"6aa12d8eedcf69cf2235178b","author":{"_id":"6aa12ca87d9f4226e31173ec","avatarUrl":"/avatars/83deea16c360fdaef6248dfe0f08885d.svg","fullname":"3 patti land","name":"3pattiland","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-09-09T09:57:34.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"NeoHorse-1 offers an interesting look at how agentic systems can improve through recursive learning and better routing. In gaming, 3 Patti Land also focuses on creating a smooth and engaging experience for card-game players. Those interested in 3 Patti Land can visit https://3patti-land.pk/ to explore the game and its features.\n","html":"<p>NeoHorse-1 offers an interesting look at how agentic systems can improve through recursive learning and better routing. In gaming, 3 Patti Land also focuses on creating a smooth and engaging experience for card-game players. Those interested in 3 Patti Land can visit <a href=\"https://3patti-land.pk/\" rel=\"nofollow\">https://3patti-land.pk/</a> to explore the game and its features.</p>\n","updatedAt":"2026-09-09T09:57:34.305Z","author":{"_id":"6aa12ca87d9f4226e31173ec","avatarUrl":"/avatars/83deea16c360fdaef6248dfe0f08885d.svg","fullname":"3 patti land","name":"3pattiland","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9549528360366821},"editors":["3pattiland"],"editorAvatarUrls":["/avatars/83deea16c360fdaef6248dfe0f08885d.svg"],"reactions":[],"isReport":false}},{"id":"6aa133a02cedee3d5ad04b65","author":{"_id":"6aa12ca87d9f4226e31173ec","avatarUrl":"/avatars/83deea16c360fdaef6248dfe0f08885d.svg","fullname":"3 patti land","name":"3pattiland","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-09-09T10:23:28.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"NeoHorse-1: Towards Recursive Self-Improvement highlights an interesting approach to improving AI agents through agentic post-training and adaptive routing. In a different digital space, 3 Patti Vegas is a gaming-related website focused on providing an engaging online card-game experience for players. Those interested in 3 Patti Vegas can learn more at https://3pattivegass.pk/.\n","html":"<p>NeoHorse-1: Towards Recursive Self-Improvement highlights an interesting approach to improving AI agents through agentic post-training and adaptive routing. In a different digital space, 3 Patti Vegas is a gaming-related website focused on providing an engaging online card-game experience for players. Those interested in 3 Patti Vegas can learn more at <a href=\"https://3pattivegass.pk/\" rel=\"nofollow\">https://3pattivegass.pk/</a>.</p>\n","updatedAt":"2026-09-09T10:23:28.299Z","author":{"_id":"6aa12ca87d9f4226e31173ec","avatarUrl":"/avatars/83deea16c360fdaef6248dfe0f08885d.svg","fullname":"3 patti land","name":"3pattiland","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9275226593017578},"editors":["3pattiland"],"editorAvatarUrls":["/avatars/83deea16c360fdaef6248dfe0f08885d.svg"],"reactions":[],"isReport":false},"replies":[{"id":"6aa148042799c882043467e1","author":{"_id":"6a27a70ec151fa93b00a6eed","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/BOPPCyvxRSpMH3vJyPkzK.jpeg","fullname":"TokenRhythmAdmin","name":"TokenRhythmAdmin","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-09-09T11:50:28.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Thanks for your comment. The parallel is actually an interesting one — card games like 3 Patti and agent training share a common loop: read the table (gather evidence), adjust your strategy (update constraints), act (execute), then use the outcome as feedback. That observe–judge–act–verify cycle is exactly what Agentic Post-Training in NeoHorse-1 tries to strengthen in the model. ","html":"<p>Thanks for your comment. The parallel is actually an interesting one — card games like 3 Patti and agent training share a common loop: read the table (gather evidence), adjust your strategy (update constraints), act (execute), then use the outcome as feedback. That observe–judge–act–verify cycle is exactly what Agentic Post-Training in NeoHorse-1 tries to strengthen in the model. </p>\n","updatedAt":"2026-09-09T11:50:28.238Z","author":{"_id":"6a27a70ec151fa93b00a6eed","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/BOPPCyvxRSpMH3vJyPkzK.jpeg","fullname":"TokenRhythmAdmin","name":"TokenRhythmAdmin","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9274305701255798},"editors":["TokenRhythmAdmin"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/BOPPCyvxRSpMH3vJyPkzK.jpeg"],"reactions":[],"isReport":false,"parentCommentId":"6aa133a02cedee3d5ad04b65"}}]}],"primaryEmailConfirmed":false,"paper":{"id":"2609.08183","authors":[{"_id":"6aa0c529d0174964227bec47","name":"NeoHorse Team","hidden":false},{"_id":"6aa0c529d0174964227bec48","name":"Guoliang Cao","hidden":false},{"_id":"6aa0c529d0174964227bec49","name":"Guohao Dai","hidden":false},{"_id":"6aa0c529d0174964227bec4a","name":"Tianyu Guo","hidden":false},{"_id":"6aa0c529d0174964227bec4b","name":"Kai Han","hidden":false},{"_id":"6aa0c529d0174964227bec4c","name":"Hailin Hu","hidden":false},{"_id":"6aa0c529d0174964227bec4d","name":"Zihan Jiang","hidden":false},{"_id":"6aa0c529d0174964227bec4e","name":"Xiang Kuang","hidden":false},{"_id":"6aa0c529d0174964227bec4f","name":"Boxun Li","hidden":false},{"_id":"6aa0c529d0174964227bec50","name":"Yulong Li","hidden":false},{"_id":"6aa0c529d0174964227bec51","name":"Zehua Pei","hidden":false},{"_id":"6aa0c529d0174964227bec52","name":"Yuchuan Tian","hidden":false},{"_id":"6aa0c529d0174964227bec53","name":"Jiamin Wang","hidden":false},{"_id":"6aa0c529d0174964227bec54","name":"Yu Wang","hidden":false},{"_id":"6aa0c529d0174964227bec55","name":"Yunhe Wang","hidden":false},{"_id":"6aa0c529d0174964227bec56","name":"Yihong Wu","hidden":false},{"_id":"6aa0c529d0174964227bec57","name":"Haiyang Xu","hidden":false},{"_id":"6aa0c529d0174964227bec58","name":"Shuo Zhang","hidden":false},{"_id":"6aa0c529d0174964227bec59","name":"Hang Zhou","hidden":false},{"_id":"6aa0c529d0174964227bec5a","name":"Siyang Cheng","hidden":false},{"_id":"6aa0c529d0174964227bec5b","name":"Jiayu Fan","hidden":false},{"_id":"6aa0c529d0174964227bec5c","name":"Wei He","hidden":false},{"_id":"6aa0c529d0174964227bec5d","name":"Qingrui Jiao","hidden":false},{"_id":"6aa0c529d0174964227bec5e","name":"Hongguang Li","hidden":false},{"_id":"6aa0c529d0174964227bec5f","name":"Zhiyuan Li","hidden":false},{"_id":"6aa0c529d0174964227bec60","name":"Runke Liu","hidden":false},{"_id":"6aa0c529d0174964227bec61","name":"Xi Liu","hidden":false},{"_id":"6aa0c529d0174964227bec62","name":"Xinchen Liu","hidden":false},{"_id":"6aa0c529d0174964227bec63","name":"Sinno Jialin Pan","hidden":false},{"_id":"6aa0c529d0174964227bec64","name":"Yi Ren","hidden":false},{"_id":"6aa0c529d0174964227bec65","name":"Liuyang Song","hidden":false},{"_id":"6aa0c529d0174964227bec66","name":"Chenyu Wang","hidden":false},{"_id":"6aa0c529d0174964227bec67","name":"Bei Yu","hidden":false},{"_id":"6aa0c529d0174964227bec68","name":"Quanlu Zhang","hidden":false},{"_id":"6aa0c529d0174964227bec69","name":"Xiangyu Zhang","hidden":false},{"_id":"6aa0c529d0174964227bec6a","name":"Mengyu Zheng","hidden":false},{"_id":"6aa0c529d0174964227bec6b","name":"Yingjie Zong","hidden":false}],"publishedAt":"2026-09-08T00:00:00.000Z","submittedOnDailyAt":"2026-09-09T00:00:00.000Z","title":"NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness","submittedOnDailyBy":{"_id":"6527c063e86758eb6ca800a1","avatarUrl":"/avatars/9091be87eea518209c1de9eebfa663c0.svg","isPro":false,"fullname":"JarvisPei","user":"JarvisPei","type":"user","name":"JarvisPei"},"summary":"Recursive self-improvement (RSI) requires a concrete mechanism through which an AI system observes its capabilities and converts that evidence into the next round of learning. We present NeoHorse-1, a family of agent-native models developed to explore this path through agentic post-training. Our system combines a heterogeneous model pool with intelligent routing, recording the predicted capability demand, selected service tier, and subsequent interaction for each user turn. These records are converted into training examples that preserve interleaved reasoning, tool calls, and harness context, and are admitted through structural validation, six-dimensional semantic evaluation, and subscene-level labeling. Routing signals organize supervised fine-tuning into a three-stage curriculum and extend to routing-guided on-policy distillation, where a teacher supervises student-generated responses under the same progression. Capability-guided allocation then converts evaluation feedback into the next training mixture, closing an evaluation-selection-update loop in which what the system learns to do shapes what it learns from next. Across eleven benchmarks covering harness-based agents, tool use, coding, and instruction following, post-training raises the macro-average from 58.94 to 64.87 at 4B and from 65.60 to 69.04 at 9B, substantially narrowing the aggregate gap between the post-trained 4B model and the 9B base model. NeoHorse-1 provides an initial prototype of this feedback-driven process and a path toward harness-mediated RSI across successive iterations.","upvotes":223,"discussionId":"6aa0c529d0174964227bec6c","projectPage":"https://huggingface.co/collections/TokenRhythm/neohorse-1","githubRepo":"https://github.com/TokenRhythm/NeoHorse","githubRepoAddedBy":"user","ai_summary":"NeoHorse-1 uses agentic post-training with intelligent routing, structured feedback loops, and curriculum-based distillation to improve model capabilities across agent benchmarks.","ai_keywords":["recursive self-improvement","agent-native models","agentic post-training","intelligent routing","supervised fine-tuning","routing-guided on-policy distillation","capability-guided allocation","harness-mediated RSI"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":33,"organization":{"_id":"6a27a8a19add684d19e73d40","name":"TokenRhythm","fullname":"TokenRhythm","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a27a70ec151fa93b00a6eed/ETg9yUkTJ7XauYMd3XqSH.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6527c063e86758eb6ca800a1","avatarUrl":"/avatars/9091be87eea518209c1de9eebfa663c0.svg","isPro":false,"fullname":"JarvisPei","user":"JarvisPei","type":"user"},{"_id":"6621ccfc90e03ac080050d34","avatarUrl":"/avatars/b24bd029e8ceafe724539f569bf5486a.svg","isPro":false,"fullname":"Alex Guo","user":"AlexDL","type":"user"},{"_id":"6406ac13a577649430c61faa","avatarUrl":"/avatars/c528c3b1524f2c928bda12e92796a0f3.svg","isPro":false,"fullname":"宗英杰","user":"Zyjay","type":"user"},{"_id":"6a27a70ec151fa93b00a6eed","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/BOPPCyvxRSpMH3vJyPkzK.jpeg","isPro":false,"fullname":"TokenRhythmAdmin","user":"TokenRhythmAdmin","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"6a835c9f210c6ea8c0be0a57","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a835c9f210c6ea8c0be0a57/pT5e3TLzX6CDjrnA4jsQW.jpeg","isPro":false,"fullname":"Eric R. Fujita","user":"EricFujita","type":"user"},{"_id":"6a9d38df02406bdddb33acd8","avatarUrl":"/avatars/e2bacae7689b61c0b28807188f252876.svg","isPro":false,"fullname":"ruhook","user":"ruhook","type":"user"},{"_id":"6a9e52c7c34adc18d4f5e86a","avatarUrl":"/avatars/bd91cec7bfea2f60a7f21ae040e8ba15.svg","isPro":false,"fullname":"JiaoQingRui","user":"jiaoqingrui","type":"user"},{"_id":"6a14801df432ac02b881e091","avatarUrl":"/avatars/0cf3308977f6066b02e083598cb4726f.svg","isPro":false,"fullname":"Julian Perez","user":"julianpere32","type":"user"},{"_id":"6a159353fb7af3d146ab8190","avatarUrl":"/avatars/bdf5cc8ccf06ce7b735c83adc012b09d.svg","isPro":false,"fullname":"Aiden Williams","user":"aiden-williams","type":"user"},{"_id":"63ac5701c21e60a3e9b58aa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63ac5701c21e60a3e9b58aa7/g6EX7diOpuA94R2ab-rZC.png","isPro":true,"fullname":"Dipankar Sarkar","user":"dipankarsarkar","type":"user"},{"_id":"6a147f4f695d577a5249a9c8","avatarUrl":"/avatars/1194f0e6d9c809dd7767c413d64cd889.svg","isPro":false,"fullname":"Emily Brown","user":"emily-brown2025","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":1,"organization":{"_id":"6a27a8a19add684d19e73d40","name":"TokenRhythm","fullname":"TokenRhythm","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/6a27a70ec151fa93b00a6eed/ETg9yUkTJ7XauYMd3XqSH.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2609/2609.08183.md","query":{}}">