<a href=\"https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/zcgMdV0473OVIMPAv_jGT.png\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/60f1abe7544c2adfd699860c/zcgMdV0473OVIMPAv_jGT.png\" alt=\"Screenshot 2026-05-20 at 8.54.48 AM\"></a></p>\n","updatedAt":"2026-05-20T12:55:08.297Z","author":{"_id":"60f1abe7544c2adfd699860c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674929746905-60f1abe7544c2adfd699860c.jpeg","fullname":"AK","name":"akhaliq","type":"user","isPro":false,"isHf":true,"isHfAdmin":false,"isMod":false,"followerCount":9548,"isUserFollowing":false,"primaryOrg":{"avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1583856921041-5dd96eb166059660ed1ee413.png","fullname":"Hugging Face","name":"huggingface","type":"org","isHf":true,"details":"The AI community building the future.","plan":"team"}}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.49690473079681396},"editors":["akhaliq"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/1674929746905-60f1abe7544c2adfd699860c.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.19633","authors":[{"_id":"6a0daf054cbad04167ffa4ec","name":"Lakshya A Agrawal","hidden":false},{"_id":"6a0daf054cbad04167ffa4ed","name":"Donghyun Lee","hidden":false},{"_id":"6a0daf054cbad04167ffa4ee","name":"Shangyin Tan","hidden":false},{"_id":"6a0daf054cbad04167ffa4ef","name":"Wenjie Ma","hidden":false},{"_id":"6a0daf054cbad04167ffa4f0","name":"Karim Elmaaroufi","hidden":false},{"_id":"6a0daf054cbad04167ffa4f1","name":"Rohit Sandadi","hidden":false},{"_id":"6a0daf054cbad04167ffa4f2","name":"Sanjit A. Seshia","hidden":false},{"_id":"6a0daf054cbad04167ffa4f3","name":"Koushik Sen","hidden":false},{"_id":"6a0daf054cbad04167ffa4f4","name":"Dan Klein","hidden":false},{"_id":"6a0daf054cbad04167ffa4f5","name":"Ion Stoica","hidden":false},{"_id":"6a0daf054cbad04167ffa4f6","name":"Joseph E. Gonzalez","hidden":false},{"_id":"6a0daf054cbad04167ffa4f7","name":"Omar Khattab","hidden":false},{"_id":"6a0daf054cbad04167ffa4f8","name":"Alexandros G. Dimakis","hidden":false},{"_id":"6a0daf054cbad04167ffa4f9","name":"Matei Zaharia","hidden":false}],"publishedAt":"2026-05-19T00:00:00.000Z","submittedOnDailyAt":"2026-05-20T00:00:00.000Z","title":"optimize_anything: A Universal API for Optimizing any Text Parameter","submittedOnDailyBy":{"_id":"60f1abe7544c2adfd699860c","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/1674929746905-60f1abe7544c2adfd699860c.jpeg","isPro":false,"fullname":"AK","user":"akhaliq","type":"user","name":"akhaliq"},"summary":"Can a single LLM-based optimization system match specialized tools across fundamentally different domains? We show that when optimization problems are formulated as improving a text artifact evaluated by a scoring function, a single AI-based optimization system-supporting single-task search, multi-task search with cross-problem transfer, and generalization to unseen inputs-achieves state-of-the-art results across six diverse tasks. Our system discovers agent architectures that nearly triple Gemini Flash's ARC-AGI accuracy (32.5% to 89.5%), finds scheduling algorithms that cut cloud costs by 40%, generates CUDA kernels where 87% match or beat PyTorch, and outperforms AlphaEvolve's reported circle packing solution (n=26). Ablations across three domains reveal that actionable side information yields faster convergence and substantially higher final scores than score-only feedback, and that multi-task search outperforms independent optimization given equivalent per-problem budget through cross-task transfer, with benefits scaling with the number of related tasks. Together, we show for the first time that text optimization with LLM-based search is a general-purpose problem-solving paradigm, unifying tasks traditionally requiring domain-specific algorithms under a single framework. We open-source optimize\\_anything with support for multiple backends as part of the GEPA project at https://github.com/gepa-ai/gepa .","upvotes":0,"discussionId":"6a0daf054cbad04167ffa4fa","projectPage":"https://gepa-ai.github.io/gepa/blog/2026/02/18/introducing-optimize-anything/","ai_summary":"A single LLM-based optimization system demonstrates state-of-the-art performance across diverse domains by formulating optimization problems as text artifact improvement with scoring functions, achieving superior results in AI agent discovery, cloud scheduling, CUDA kernel generation, and geometric packing compared to specialized tools.","ai_keywords":["LLM-based optimization system","text artifact evaluation","scoring function","multi-task search","cross-problem transfer","generalization","actionaable side information","convergence","cross-task transfer"]},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[],"acceptLanguages":["en"],"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.19633.md"}">
optimize_anything: A Universal API for Optimizing any Text Parameter
Published on May 19
· Submitted by AK on May 20 Authors: ,
,
,
,
,
,
,
,
,
,
,
,
,
Abstract
A single LLM-based optimization system demonstrates state-of-the-art performance across diverse domains by formulating optimization problems as text artifact improvement with scoring functions, achieving superior results in AI agent discovery, cloud scheduling, CUDA kernel generation, and geometric packing compared to specialized tools.
AI-generated summary
Can a single LLM-based optimization system match specialized tools across fundamentally different domains? We show that when optimization problems are formulated as improving a text artifact evaluated by a scoring function, a single AI-based optimization system-supporting single-task search, multi-task search with cross-problem transfer, and generalization to unseen inputs-achieves state-of-the-art results across six diverse tasks. Our system discovers agent architectures that nearly triple Gemini Flash's ARC-AGI accuracy (32.5% to 89.5%), finds scheduling algorithms that cut cloud costs by 40%, generates CUDA kernels where 87% match or beat PyTorch, and outperforms AlphaEvolve's reported circle packing solution (n=26). Ablations across three domains reveal that actionable side information yields faster convergence and substantially higher final scores than score-only feedback, and that multi-task search outperforms independent optimization given equivalent per-problem budget through cross-task transfer, with benefits scaling with the number of related tasks. Together, we show for the first time that text optimization with LLM-based search is a general-purpose problem-solving paradigm, unifying tasks traditionally requiring domain-specific algorithms under a single framework. We open-source optimize\_anything with support for multiple backends as part of the GEPA project at https://github.com/gepa-ai/gepa .
Community
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/2605.19633 in a model README.md to link it from this page.
Cite arxiv.org/abs/2605.19633 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2605.19633 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.