LongHorizon-Harness: Advancing Long-Horizon Agents for Real-World Tasks<br>Github: <a href=\"https://github.com/AMAP-ML/LongHorizon-Harness\" rel=\"nofollow\">https://github.com/AMAP-ML/LongHorizon-Harness</a><br>Website: <a href=\"https://lh-harness.pages.dev\" rel=\"nofollow\">https://lh-harness.pages.dev</a></p>\n","updatedAt":"2026-08-04T05:56:46.841Z","author":{"_id":"64d1dc5273174cecdffc97d3","avatarUrl":"/avatars/6564e6b68fee9673f75b6366adf39a3b.svg","fullname":"Wang Yong","name":"seashell11","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.6997005939483643},"editors":["seashell11"],"editorAvatarUrls":["/avatars/6564e6b68fee9673f75b6366adf39a3b.svg"],"reactions":[{"reaction":"👍","users":["xiaochonglinghu","poiuytrewq123","lerogo"],"count":3}],"isReport":false}},{"id":"6a718d9585271b1cc97346f5","author":{"_id":"6a718d5a544004d9e6db37d7","avatarUrl":"/avatars/70d162d329b57c1af513ac633e267a41.svg","fullname":"natasha gandhi","name":"natasha052gandhi","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false},"createdAt":"2026-08-04T06:58:29.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Interesting approach! Separating task state from execution and verifying it independently seems like a practical way to improve reliability for long-horizon AI agents. The benchmark improvements are impressive.","html":"<p>Interesting approach! Separating task state from execution and verifying it independently seems like a practical way to improve reliability for long-horizon AI agents. The benchmark improvements are impressive.</p>\n","updatedAt":"2026-08-04T06:58:29.458Z","author":{"_id":"6a718d5a544004d9e6db37d7","avatarUrl":"/avatars/70d162d329b57c1af513ac633e267a41.svg","fullname":"natasha gandhi","name":"natasha052gandhi","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.9262194037437439},"editors":["natasha052gandhi"],"editorAvatarUrls":["/avatars/70d162d329b57c1af513ac633e267a41.svg"],"reactions":[{"reaction":"👍","users":["poiuytrewq123","lerogo"],"count":2}],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.01964","authors":[{"_id":"6a715744ec5082b9f872ce05","name":"Ziyu Ma","hidden":false},{"_id":"6a715744ec5082b9f872ce06","name":"Hailang Huang","hidden":false},{"_id":"6a715744ec5082b9f872ce07","name":"Shun Zou","hidden":false},{"_id":"6a715744ec5082b9f872ce08","name":"Yong Wang","hidden":false},{"_id":"6a715744ec5082b9f872ce09","name":"Shidong Yang","hidden":false},{"_id":"6a715744ec5082b9f872ce0a","name":"Yiming Hu","hidden":false},{"_id":"6a715744ec5082b9f872ce0b","name":"Fei Wei","hidden":false},{"_id":"6a715744ec5082b9f872ce0c","name":"XiangXiang Chu","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/64d1dc5273174cecdffc97d3/QDDq6oSuxJP_g4SwESa7B.mp4"],"publishedAt":"2026-08-03T00:00:00.000Z","submittedOnDailyAt":"2026-08-04T00:00:00.000Z","title":"LongHorizon-Harness: Advancing Long-Horizon Agents for Real-World Tasks","submittedOnDailyBy":{"_id":"64d1dc5273174cecdffc97d3","avatarUrl":"/avatars/6564e6b68fee9673f75b6366adf39a3b.svg","isPro":false,"fullname":"Wang Yong","user":"seashell11","type":"user","name":"seashell11"},"summary":"Large language model (LLM) agents increasingly undertake long-horizon tasks that require sustained reasoning, tool use, and revision across many interdependent steps. However, existing agent harnesses maintain task execution, task state, and completion assessment within a growing context, making the state difficult to track and allowing incorrect self-assessments to propagate into later decisions. We reformulate long-horizon execution as a task-state management problem and propose LongHorizon-Harness, which maintains the task state explicitly outside execution and updates it only with facts independently verified from the environment. Its Manage-Execute-Audit(MEA) loop uses a manager to maintain the task state and determine the next subtask, a fresh-context executor to perform it, and a read-only auditor to verify the resulting environment state before the next round. A lightweight AgentAdapter supports interchangeable model and harness backends without modifying their native agent loops. LongHorizon-Harness improves Qwen~3.7-Plus from 51.8% to 80.7% on WeaveBench, from 69.7% to 77.2% on Terminal-Bench~2.1, and from 2.8% to 8.3% on OSWorld~2.0. It also raises Claude Opus~4.7 from 20.0% to 34.3% on an OSWorld2.0 subset, demonstrating consistent gains across models, harnesses, and interaction domains.","upvotes":67,"discussionId":"6a715745ec5082b9f872ce0d","projectPage":"https://lh-harness.pages.dev","githubRepo":"https://github.com/AMAP-ML/LongHorizon-Harness","githubRepoAddedBy":"user","githubStars":51},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64d1dc5273174cecdffc97d3","avatarUrl":"/avatars/6564e6b68fee9673f75b6366adf39a3b.svg","isPro":false,"fullname":"Wang Yong","user":"seashell11","type":"user"},{"_id":"65003db8bef9b594656f8fa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/65003db8bef9b594656f8fa7/L6cvPOAeBRnFnIQwWxYyf.png","isPro":false,"fullname":"Hailang Huang","user":"lerogo","type":"user"},{"_id":"6682775501c30ad93ec5e500","avatarUrl":"/avatars/971ee2028589f6089559306b40a58da0.svg","isPro":false,"fullname":"Jiashu Zhu","user":"Jiashuz","type":"user"},{"_id":"6773bcaa675a971ddf1e81dd","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/a8VUwZYXd7O_mq_zFvXMh.png","isPro":false,"fullname":"CokeWang","user":"CokeWang","type":"user"},{"_id":"6964bd185b2c1aaa90b6dc3d","avatarUrl":"/avatars/6a9e4159e4db844115c3664e71b34ce6.svg","isPro":false,"fullname":"douhaoexia","user":"douhaoexia","type":"user"},{"_id":"697c619099afe12fc887f046","avatarUrl":"/avatars/9682ef176a74c5d2e2e4dec2427a0fea.svg","isPro":false,"fullname":"Zesheng wang","user":"Brucelee0118","type":"user"},{"_id":"6465e8abe03f25ce60f5799f","avatarUrl":"/avatars/21b2b4b2688e59bb1dfca2591cdb2d23.svg","isPro":false,"fullname":"renda Li","user":"SSSAMMMM","type":"user"},{"_id":"6964bb636e44e2b2e26c418d","avatarUrl":"/avatars/bed542d8f428e957f299e8d8a89c0145.svg","isPro":false,"fullname":"HotTricker","user":"yang8748","type":"user"},{"_id":"661de9defdbc9c247f159d15","avatarUrl":"/avatars/38e21e78327cc908201122405c48f41b.svg","isPro":false,"fullname":"Rui Dai","user":"DerryD","type":"user"},{"_id":"6964a33d2ab4d967d0543d84","avatarUrl":"/avatars/93702bd3303bccb4ee29812e3025e7aa.svg","isPro":false,"fullname":"liuzheng","user":"lzlz2000","type":"user"},{"_id":"6901b520f8f20b9d7015a38d","avatarUrl":"/avatars/49e8d709818d1f0778758e39bd39f0a3.svg","isPro":false,"fullname":"zoushun","user":"shunzou1314","type":"user"},{"_id":"6459f2ae896f285ceb2384f0","avatarUrl":"/avatars/6bf4ed9ddd8f4dd45a97bec29274ae38.svg","isPro":false,"fullname":"wf","user":"Olivia0","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":2,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.01964.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2608.01964 in a model README.md to link it from this page.

Cite arxiv.org/abs/2608.01964 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2608.01964 in a Space README.md to link it from this page.

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