Recursive self-improvement (RSI) requires AI systems that improve the process of building AI (i.e., AI4AI); machine learning engineering (MLE) offers a concrete, executable testbed for studying this capability. We introduce OpenMLE, an open full-stack system for RSI research in MLE, and Frontis-MA1 (35B) as a meta-evolution agent for MLE, exceeding GPT-5.5 + Codex and approaching GPT-5.6 Sol and the 2.8T Kimi K3 on MLE-Bench Lite with fixed budget.</p>\n","updatedAt":"2026-07-31T02:25:00.409Z","author":{"_id":"60bc94cd85a3ab33829b6211","avatarUrl":"/avatars/b57d36c7577fbbb42ea5b963eef4144a.svg","fullname":"Kaiyan Zhang","name":"iseesaw","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":13,"isUserFollowing":false}},"numEdits":2,"identifiedLanguage":{"language":"en","probability":0.46662837266921997},"editors":["iseesaw"],"editorAvatarUrls":["/avatars/b57d36c7577fbbb42ea5b963eef4144a.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2607.28568","authors":[{"_id":"6a6c05ab7bd25d8874c0712c","name":"Junlin Yang","hidden":false},{"_id":"6a6c05ab7bd25d8874c0712d","name":"Che Jiang","hidden":false},{"_id":"6a6c05ab7bd25d8874c0712e","name":"Yu Fu","hidden":false},{"_id":"6a6c05ab7bd25d8874c0712f","name":"Tianwei Luo","hidden":false},{"_id":"6a6c05ab7bd25d8874c07130","name":"Can Ren","hidden":false},{"_id":"6a6c05ab7bd25d8874c07131","name":"Weizhi Wang","hidden":false},{"_id":"6a6c05ab7bd25d8874c07132","name":"Kaikai Zhao","hidden":false},{"_id":"6a6c05ab7bd25d8874c07133","user":{"_id":"66d12f904cc92b9866c7f837","avatarUrl":"/avatars/5c3ce46c1f953a7081e344cd91a8d17b.svg","isPro":false,"fullname":"Hongyi Liu","user":"hongyi-liu","type":"user","name":"hongyi-liu"},"name":"Hongyi Liu","status":"claimed_verified","statusLastChangedAt":"2026-07-31T08:45:05.499Z","hidden":false},{"_id":"6a6c05ab7bd25d8874c07134","name":"Yuxin Zuo","hidden":false},{"_id":"6a6c05ab7bd25d8874c07135","name":"Yuru Wang","hidden":false},{"_id":"6a6c05ab7bd25d8874c07136","name":"Yuchen Fan","hidden":false},{"_id":"6a6c05ab7bd25d8874c07137","name":"Kai Tian","hidden":false},{"_id":"6a6c05ab7bd25d8874c07138","name":"Zhenzhao Yuan","hidden":false},{"_id":"6a6c05ab7bd25d8874c07139","name":"Xiaojian Lin","hidden":false},{"_id":"6a6c05ab7bd25d8874c0713a","name":"Li Sheng","hidden":false},{"_id":"6a6c05ab7bd25d8874c0713b","name":"Rushi Qiang","hidden":false},{"_id":"6a6c05ab7bd25d8874c0713c","name":"Guoli Jia","hidden":false},{"_id":"6a6c05ab7bd25d8874c0713d","name":"Xingtai Lv","hidden":false},{"_id":"6a6c05ab7bd25d8874c0713e","name":"Ermo Hua","hidden":false},{"_id":"6a6c05ab7bd25d8874c0713f","name":"Dianqiao Lei","hidden":false},{"_id":"6a6c05ab7bd25d8874c07140","name":"Youbang Sun","hidden":false},{"_id":"6a6c05ab7bd25d8874c07141","name":"Ning Ding","hidden":false},{"_id":"6a6c05ab7bd25d8874c07142","name":"Bowen Zhou","hidden":false},{"_id":"6a6c05ab7bd25d8874c07143","user":{"_id":"60bc94cd85a3ab33829b6211","avatarUrl":"/avatars/b57d36c7577fbbb42ea5b963eef4144a.svg","isPro":false,"fullname":"Kaiyan Zhang","user":"iseesaw","type":"user","name":"iseesaw"},"name":"Kaiyan Zhang","status":"claimed_verified","statusLastChangedAt":"2026-07-31T08:45:05.492Z","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/60bc94cd85a3ab33829b6211/W7-x4amh4s7CM-B6OPpmJ.mp4"],"publishedAt":"2026-07-30T00:00:00.000Z","submittedOnDailyAt":"2026-07-31T00:00:00.000Z","title":"Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering","submittedOnDailyBy":{"_id":"60bc94cd85a3ab33829b6211","avatarUrl":"/avatars/b57d36c7577fbbb42ea5b963eef4144a.svg","isPro":false,"fullname":"Kaiyan Zhang","user":"iseesaw","type":"user","name":"iseesaw"},"summary":"Recursive self-improvement (RSI) requires AI systems that improve the process of building AI (i.e., AI4AI); machine learning engineering (MLE) offers a concrete, executable testbed for studying this capability. We introduce OpenMLE, an open full-stack system for RSI research in MLE, spanning verifiable task environments with execution feedback (OpenMLE-Gym), operator learning (OpenMLE-RL), and long-horizon search (OpenMLE-Evo). On this stack we post-train Frontis-MA1 (35B) as a meta-evolution agent for MLE, aligning post-training and inference around four atomic program-evolution operators (Draft, Improve, Debug, Crossover): the same operators are trained via execution-grounded SFT and RL on data deduplicated against all evaluation benchmarks, then composed into long-horizon search, coupling learning and evolution in a single loop. On MLE-Bench Lite under a 12-hour per-task budget on one RTX 4090 capped at 12 GB VRAM, Frontis-MA1 (35B) improves Medal Average from 39.39% to 60.61% over its base model with OpenMLE-Evo, and reaches 71.21% with OpenMLE-Evo-Max (benchmark-independent experience priors and asynchronous search), exceeding GPT-5.5 + Codex and approaching GPT-5.6 Sol and the 2.8T Kimi K3. On held-out NatureBench Lite, both components transfer: with the framework fixed, swapping in the trained model raises Match-SOTA from 50% to 70%; with the model fixed, swapping in OpenMLE-Evo raises it from 20% to 50%. We release the model weights and the full OpenMLE stack to enable reproducible research on executable AI4AI toward RSI. Code: https://github.com/FrontisAI/OpenRSI","upvotes":111,"discussionId":"6a6c05ab7bd25d8874c07144","projectPage":"https://frontisai.github.io/OpenRSI/","githubRepo":"https://github.com/FrontisAI/OpenRSI","githubRepoAddedBy":"user","githubStars":1,"organization":{"_id":"6a32950e4b5c1c0ebee0e552","name":"FrontisAI","fullname":"Frontis AI","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/60bc94cd85a3ab33829b6211/1w_MutesbGw4NwNkA_dn5.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"60bc94cd85a3ab33829b6211","avatarUrl":"/avatars/b57d36c7577fbbb42ea5b963eef4144a.svg","isPro":false,"fullname":"Kaiyan Zhang","user":"iseesaw","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":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"66ed083acaf696884760729a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/RgPe99BqsJsHUWoXO1qtS.jpeg","isPro":false,"fullname":"Nick Yang","user":"RadioBlue","type":"user"},{"_id":"675aa66a8ff80593f8214c3e","avatarUrl":"/avatars/ae90502e36f061692f59193f0da22e3d.svg","isPro":false,"fullname":"abab112","user":"abab112","type":"user"},{"_id":"6a3a0be66bc9e697359f9fef","avatarUrl":"/avatars/3f84fa43531d7924c6a257d6e6cab0db.svg","isPro":false,"fullname":"Weizhi Wang","user":"wangweiz22","type":"user"},{"_id":"6a3a481f96be7e84d2a48ec2","avatarUrl":"/avatars/44b741efac70fcf86597fbf89293954a.svg","isPro":false,"fullname":"jimmy95200","user":"jimmy95200","type":"user"},{"_id":"68961040cd4db3f88477ba78","avatarUrl":"/avatars/6a60246b35b885ad04da4b4c488ee68a.svg","isPro":false,"fullname":"Dianqiao Lei","user":"DianQiao","type":"user"},{"_id":"662de78652e194d5d4b63d18","avatarUrl":"/avatars/3d74efd07258a7a8146ee673d752f9c8.svg","isPro":false,"fullname":"kuo","user":"zhangkuo2024","type":"user"},{"_id":"686b31ea9c4cd7445680095a","avatarUrl":"/avatars/6edfb953556da6300850207e4fd6e436.svg","isPro":false,"fullname":"aerway","user":"aerway","type":"user"},{"_id":"6a3b9ac24669267041f6378f","avatarUrl":"/avatars/3275d0ec097be583ca5c9f1a31759645.svg","isPro":false,"fullname":"exile","user":"lizhuorui","type":"user"},{"_id":"64802b46c57f629056c578ee","avatarUrl":"/avatars/50748f7b782c763a23e4bf04869a3466.svg","isPro":false,"fullname":"yiyi","user":"cnwang","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":2,"organization":{"_id":"6a32950e4b5c1c0ebee0e552","name":"FrontisAI","fullname":"Frontis AI","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/60bc94cd85a3ab33829b6211/1w_MutesbGw4NwNkA_dn5.jpeg"},"query":{}}">

Cite arxiv.org/abs/2607.28568 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.