Excited to share Mendel Gödel Machine (MGM)! 🧬</p>\n<p>Recent self-evolving agents have shown that agents can improve themselves through iterative evaluation and modification. But in many existing systems, each evolution step is still driven by only a single evaluation trajectory, while a large amount of informative experience generated during evaluation is left unused. At the same time, tree-based evolution naturally organizes agents into separate lineages, making it difficult for valuable discoveries from one branch to benefit others.</p>\n<p>MGM is motivated by a simple idea: self-evolution should learn from the broader evolutionary process, not just from one local failure. By better exploiting comparative signals across evaluation trajectories and enabling information transfer across lineages, we aim to make each evolution step more informed, efficient, and reliable.</p>\n<p>Our key question is no longer just “Can agents evolve?”, but “How can they evolve with higher quality—and ultimately scale self-evolution?”</p>\n","updatedAt":"2026-08-12T02:47:40.794Z","author":{"_id":"69551530b383af0a6c2f9d12","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/69551530b383af0a6c2f9d12/gFoSZ7xW3wbvCxzd_T-Qx.jpeg","fullname":"Changzhi Liu","name":"RealLiu","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9287121295928955},"editors":["RealLiu"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/69551530b383af0a6c2f9d12/gFoSZ7xW3wbvCxzd_T-Qx.jpeg"],"reactions":[{"reaction":"🔥","users":["liuyilun2000","RealLiu"],"count":2}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.07645","authors":[{"_id":"6a7a986d019ce76dc7b3aa8a","user":{"_id":"69551530b383af0a6c2f9d12","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/69551530b383af0a6c2f9d12/gFoSZ7xW3wbvCxzd_T-Qx.jpeg","isPro":false,"fullname":"Changzhi Liu","user":"RealLiu","type":"user","name":"RealLiu"},"name":"Changzhi Liu","status":"claimed_verified","statusLastChangedAt":"2026-08-11T08:45:04.500Z","hidden":false},{"_id":"6a7a986d019ce76dc7b3aa8b","user":{"_id":"6380991c5c62156ce7dfae8c","avatarUrl":"/avatars/ca5186b7bcf3db9dd2db575ffdec97f8.svg","isPro":false,"fullname":"Yilun Liu","user":"liuyilun2000","type":"user","name":"liuyilun2000"},"name":"Yilun Liu","status":"claimed_verified","statusLastChangedAt":"2026-08-12T08:45:05.031Z","hidden":false},{"_id":"6a7a986d019ce76dc7b3aa8c","name":"Sikuan Yan","hidden":false},{"_id":"6a7a986d019ce76dc7b3aa8d","name":"Volker Tresp","hidden":false},{"_id":"6a7a986d019ce76dc7b3aa8e","name":"Yunpu Ma","hidden":false}],"publishedAt":"2026-08-07T00:00:00.000Z","submittedOnDailyAt":"2026-08-12T00:00:00.000Z","title":"Mendel Gödel Machine: Recursive Self-Improving Coding Agents via Comparative Evolution","submittedOnDailyBy":{"_id":"69551530b383af0a6c2f9d12","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/69551530b383af0a6c2f9d12/gFoSZ7xW3wbvCxzd_T-Qx.jpeg","isPro":false,"fullname":"Changzhi Liu","user":"RealLiu","type":"user","name":"RealLiu"},"summary":"Self-improving coding agents that iteratively rewrite their own source code have demonstrated impressive performance on coding tasks. However, existing solutions generally derive self-modification from a single failure trajectory at a time, overlooking rich comparative signals available in the agent's expanding archive of past attempts. According to Mendelian principles of controlled inheritance, we introduce Mendel Gödel Machine (MGM). In addition to the general single-trajectory clonal mutation, MGM includes two new types of self-modification that better utilizes evidences accumulated: the reaction-norm mutation edits an agent based on its trajectories on multiple tasks simultaneously, and the cross-lineage hybridization edits an agent using the trajectory of a reference agent from another lineage on the same task. Under an additive fitness landscape model, we prove theoretically and demonstrate via controlled surrogate simulation that the new strategies facilitate a faster and better convergence over single-trajectory baselines. Experiments on SWE-bench and Polyglot confirm MGM's consistent improvement in performance, efficiency, and generalizability.","upvotes":15,"discussionId":"6a7a986e019ce76dc7b3aa8f","projectPage":"https://reallcz.github.io/MGM/","githubRepo":"https://github.com/RealLcz/MGM","githubRepoAddedBy":"user","ai_summary":"Mendel Gödel Machine improves self-improving coding agents by using multi-trajectory mutations and cross-lineage hybridization to accelerate convergence and boost performance.","ai_keywords":["Mendel Gödel Machine","reaction-norm mutation","cross-lineage hybridization","self-modification","additive fitness landscape","clonal mutation"],"ai_summary_model":"thinkingmachines/Inkling-Small","githubStars":3,"organization":{"_id":"62e50495ae9d3f10acb6a9ca","name":"LMU","fullname":"Ludwig Maximilian University of Munich","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1659176121442-5fcaabed246881afd5b00167.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"69551530b383af0a6c2f9d12","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/69551530b383af0a6c2f9d12/gFoSZ7xW3wbvCxzd_T-Qx.jpeg","isPro":false,"fullname":"Changzhi Liu","user":"RealLiu","type":"user"},{"_id":"6380991c5c62156ce7dfae8c","avatarUrl":"/avatars/ca5186b7bcf3db9dd2db575ffdec97f8.svg","isPro":false,"fullname":"Yilun Liu","user":"liuyilun2000","type":"user"},{"_id":"65244bd306e839c55403beeb","avatarUrl":"/avatars/2e475eaf9885bf579ca5f791704c0a3d.svg","isPro":false,"fullname":"Difan Jiao","user":"difanjiao","type":"user"},{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user"},{"_id":"646def60df618b303b419323","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/646def60df618b303b419323/JLJGYen4-5M8ivsLsSk0w.jpeg","isPro":false,"fullname":"Lei Wang","user":"demolei","type":"user"},{"_id":"6813ddc02d792d3523017289","avatarUrl":"/avatars/f4d142bdbf788bce7914a2974a2c0fe8.svg","isPro":false,"fullname":"Ma","user":"YunpuMa","type":"user"},{"_id":"6a15f2c41140e3a5ccdfe141","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/EQymoaAoZLSxRtnKME8m5.png","isPro":false,"fullname":"高宇明","user":"levigree2","type":"user"},{"_id":"69234f5207820287f17451b3","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/rz9KC4VJ47VR9q0yVagBx.png","isPro":false,"fullname":"Stephen Moore","user":"shootstuff","type":"user"},{"_id":"69830c901da07be927e1e8c5","avatarUrl":"/avatars/1cc3d5402f22e17389bb11b60d9d0d67.svg","isPro":false,"fullname":"Marie-Claire Dupont","user":"zxcv432","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":"6a1ec930da831caff3d62c51","avatarUrl":"/avatars/b7b620526412b14af6cfb4aa6d5b7ebd.svg","isPro":false,"fullname":"Kilian Gildemeister","user":"kgildeme","type":"user"},{"_id":"62cfe10d6a61a88ea0cf88bf","avatarUrl":"/avatars/50a3a270f91a45d92e454d350e6f9af4.svg","isPro":false,"fullname":"Marco Togni","user":"KRLLRZz","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"62e50495ae9d3f10acb6a9ca","name":"LMU","fullname":"Ludwig Maximilian University of Munich","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1659176121442-5fcaabed246881afd5b00167.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.07645.md","query":{}}">
Mendel Gödel Machine: Recursive Self-Improving Coding Agents via Comparative Evolution
Abstract
Mendel Gödel Machine improves self-improving coding agents by using multi-trajectory mutations and cross-lineage hybridization to accelerate convergence and boost performance.
Self-improving coding agents that iteratively rewrite their own source code have demonstrated impressive performance on coding tasks. However, existing solutions generally derive self-modification from a single failure trajectory at a time, overlooking rich comparative signals available in the agent's expanding archive of past attempts. According to Mendelian principles of controlled inheritance, we introduce Mendel Gödel Machine (MGM). In addition to the general single-trajectory clonal mutation, MGM includes two new types of self-modification that better utilizes evidences accumulated: the reaction-norm mutation edits an agent based on its trajectories on multiple tasks simultaneously, and the cross-lineage hybridization edits an agent using the trajectory of a reference agent from another lineage on the same task. Under an additive fitness landscape model, we prove theoretically and demonstrate via controlled surrogate simulation that the new strategies facilitate a faster and better convergence over single-trajectory baselines. Experiments on SWE-bench and Polyglot confirm MGM's consistent improvement in performance, efficiency, and generalizability.
Community
This comment has been hidden (marked as Resolved) Excited to share Mendel Gödel Machine (MGM)! 🧬
Recent self-evolving agents have shown that agents can improve themselves through iterative evaluation and modification. But in many existing systems, each evolution step is still driven by only a single evaluation trajectory, while a large amount of informative experience generated during evaluation is left unused. At the same time, tree-based evolution naturally organizes agents into separate lineages, making it difficult for valuable discoveries from one branch to benefit others.
MGM is motivated by a simple idea: self-evolution should learn from the broader evolutionary process, not just from one local failure. By better exploiting comparative signals across evaluation trajectories and enabling information transfer across lineages, we aim to make each evolution step more informed, efficient, and reliable.
Our key question is no longer just “Can agents evolve?”, but “How can they evolve with higher quality—and ultimately scale self-evolution?”
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2608.07645 in a model README.md to link it from this page.
Cite arxiv.org/abs/2608.07645 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2608.07645 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.