Hugging Face Daily Papers · August 4, 2026 · 4 min read

WorldExam: Benchmarking World Models from Apparent Appearance to Inherent Reactivity

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

<a href=\"https://cdn-uploads.huggingface.co/production/uploads/66dd71d0140f04eb180d7c2a/KFrelafseOMuuxltmGsdI.png\" rel=\"nofollow\"><img src=\"https://cdn-uploads.huggingface.co/production/uploads/66dd71d0140f04eb180d7c2a/KFrelafseOMuuxltmGsdI.png\" alt=\"teaser\"></a> WorldExam is a hierarchical diagnostic benchmark for controllable video world models. Beyond visual appearance and explicit instruction following, it asks whether a model preserves a coherent world and exhibits inherent reactivity. Across camera-, action-, and language-driven interfaces, WorldExam organizes 1,474 test cases into four diagnostic levels and eight dedicated tasks under a unified evaluation protocol. Project Website: <a href=\"https://worldexam.github.io/\" rel=\"nofollow\">https://worldexam.github.io/</a>\n","updatedAt":"2026-08-04T03:48:24.384Z","author":{"_id":"66dd71d0140f04eb180d7c2a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66dd71d0140f04eb180d7c2a/9CYiiJKQ-UxxSgfjmkpoc.jpeg","fullname":"Yuxue Yang","name":"Yuppie1204","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":9,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.7288110852241516},"editors":["Yuppie1204"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/66dd71d0140f04eb180d7c2a/9CYiiJKQ-UxxSgfjmkpoc.jpeg"],"reactions":[],"isReport":false}},{"id":"6a7177efc4a24ff846716f8b","author":{"_id":"66e03eace17fb5ff054b7686","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66e03eace17fb5ff054b7686/PpSV0Qo5lwTyxIZMp57xq.jpeg","fullname":"Xiaoxi Li","name":"lixiaoxi45","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":28,"isUserFollowing":false},"createdAt":"2026-08-04T05:26:07.000Z","type":"comment","data":{"edited":false,"hidden":false,"latest":{"raw":"Congrats!","html":"Congrats!\n","updatedAt":"2026-08-04T05:26:07.109Z","author":{"_id":"66e03eace17fb5ff054b7686","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66e03eace17fb5ff054b7686/PpSV0Qo5lwTyxIZMp57xq.jpeg","fullname":"Xiaoxi Li","name":"lixiaoxi45","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":28,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.4437655210494995},"editors":["lixiaoxi45"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/66e03eace17fb5ff054b7686/PpSV0Qo5lwTyxIZMp57xq.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2608.02603","authors":[{"_id":"6a715fc1ec5082b9f872ce80","name":"Yuxue Yang","hidden":false},{"_id":"6a715fc1ec5082b9f872ce81","name":"Shuyao Shang","hidden":false},{"_id":"6a715fc1ec5082b9f872ce82","name":"Jiahe Wang","hidden":false},{"_id":"6a715fc1ec5082b9f872ce83","name":"Zitong Zhou","hidden":false},{"_id":"6a715fc1ec5082b9f872ce84","name":"Liang Tan","hidden":false},{"_id":"6a715fc1ec5082b9f872ce85","name":"Junhan Zeng","hidden":false},{"_id":"6a715fc1ec5082b9f872ce86","name":"Ruizhi Li","hidden":false},{"_id":"6a715fc1ec5082b9f872ce87","name":"Junyan Li","hidden":false},{"_id":"6a715fc1ec5082b9f872ce88","name":"Yu Liu","hidden":false},{"_id":"6a715fc1ec5082b9f872ce89","name":"Xiao Yang","hidden":false},{"_id":"6a715fc1ec5082b9f872ce8a","name":"Yong Li","hidden":false},{"_id":"6a715fc1ec5082b9f872ce8b","name":"Jun Zhu","hidden":false},{"_id":"6a715fc1ec5082b9f872ce8c","name":"Hongsheng Li","hidden":false},{"_id":"6a715fc1ec5082b9f872ce8d","name":"Tieniu Tan","hidden":false},{"_id":"6a715fc1ec5082b9f872ce8e","name":"Lue Fan","hidden":false},{"_id":"6a715fc1ec5082b9f872ce8f","name":"Zhaoxiang Zhang","hidden":false}],"mediaUrls":["https://cdn-uploads.huggingface.co/production/uploads/66dd71d0140f04eb180d7c2a/ELJM5lydsGelb63-7Df_K.png"],"publishedAt":"2026-08-03T00:00:00.000Z","submittedOnDailyAt":"2026-08-04T00:00:00.000Z","title":"WorldExam: Benchmarking World Models from Apparent Appearance to Inherent Reactivity","submittedOnDailyBy":{"_id":"66dd71d0140f04eb180d7c2a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66dd71d0140f04eb180d7c2a/9CYiiJKQ-UxxSgfjmkpoc.jpeg","isPro":false,"fullname":"Yuxue Yang","user":"Yuppie1204","type":"user","name":"Yuppie1204"},"summary":"Controllable video generation models are increasingly being developed as world models. Accordingly, evaluating them in this role extends beyond the apparent appearance of generated videos to the inherent reactivity of the worlds they depict: the ability to infer from the scene state how the world should react and to generate plausible consequences not explicitly described in the input. Yet existing benchmarks mainly assess visual quality or explicit instruction fulfillment by checking whether requested actions and interaction outcomes are realized, leaving inherent reactivity underexamined. We introduce WorldExam, a hierarchical diagnostic benchmark spanning four levels: Visual Quality, Control Adherence, Spatial Consistency, and World Reactivity. It comprises 1,474 cases across eight dedicated tasks and supports unified evaluation of camera-, action-, and language-driven model paradigms. The World Reactivity level evaluates scene-conditioned reactions and goal-directed behaviors beyond what is explicitly specified in the input. Evaluation of 20 representative models reveals a clear capability split. Camera-driven models excel at camera control, but their interfaces do not support dynamic interaction; action-driven models control subjects more precisely but often leave the world unresponsive; and language-driven models perform better on interaction but follow complex controls less faithfully. No model combines broad task coverage with consistently strong performance, showing that high visual quality and explicit instruction fulfillment do not guarantee inherent reactivity.","upvotes":18,"discussionId":"6a715fc2ec5082b9f872ce90","projectPage":"https://worldexam.github.io/","githubRepo":"https://github.com/YuxueYang1204/worldexam","githubRepoAddedBy":"user","githubStars":6},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"66dd71d0140f04eb180d7c2a","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66dd71d0140f04eb180d7c2a/9CYiiJKQ-UxxSgfjmkpoc.jpeg","isPro":false,"fullname":"Yuxue Yang","user":"Yuppie1204","type":"user"},{"_id":"6a0ab302145b4de24ae1703b","avatarUrl":"/avatars/6a97a67d6710f39afd0488f924afea01.svg","isPro":false,"fullname":"ljr","user":"Luojr","type":"user"},{"_id":"66ebb5172eb421192b4e1afb","avatarUrl":"/avatars/1fc1e196562693fd4e1b834642f54429.svg","isPro":false,"fullname":"Shuyao Shang","user":"ShuyaoShang","type":"user"},{"_id":"63c1699e40a26dd2db32400d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63c1699e40a26dd2db32400d/3N0-Zp8igv8-52mXAdiiq.jpeg","isPro":false,"fullname":"Chroma","user":"Chroma111","type":"user"},{"_id":"69abe00ceb2b5f5062d5583b","avatarUrl":"/avatars/6a833341f46f8044cb201e7f22fac89a.svg","isPro":false,"fullname":"Junhan Zeng","user":"SchNaPPiP","type":"user"},{"_id":"695b608a093cbcbd1616bfd5","avatarUrl":"/avatars/5dbc4a8b32e412f06baced4e3ad882c5.svg","isPro":false,"fullname":"Zitong Zhou","user":"potternum","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":"6a0bbe2be0287b4e9466dcef","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/_gz4EhY0Tv7ZAaN0WymqJ.png","isPro":false,"fullname":"DexJoCo","user":"DexJoCo","type":"user"},{"_id":"6642e5c8bedb81171f1f4386","avatarUrl":"/avatars/5a9675e057d4eecc9cb30828f435c8b2.svg","isPro":false,"fullname":"Liu Xueyi","user":"LiuxyIA","type":"user"},{"_id":"644cc2c36dfd5f8240d76a52","avatarUrl":"/avatars/dcd9279af1c6d8535e48dc6e3e6511cd.svg","isPro":false,"fullname":"Ziqi Shi","user":"renshengjihe","type":"user"},{"_id":"68dbbede1a55bacf3e75d58a","avatarUrl":"/avatars/ea3b540a02d10125422e87dee0ed8e6f.svg","isPro":false,"fullname":"ly","user":"ly152","type":"user"},{"_id":"66e03eace17fb5ff054b7686","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/66e03eace17fb5ff054b7686/PpSV0Qo5lwTyxIZMp57xq.jpeg","isPro":false,"fullname":"Xiaoxi Li","user":"lixiaoxi45","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2608/2608.02603.md","query":{}}">

Papers

arxiv:2608.02603

WorldExam: Benchmarking World Models from Apparent Appearance to Inherent Reactivity

Published on Aug 3

· Submitted by

Yuxue Yang on Aug 4

Upvote

Authors:

Abstract

Controllable video generation models are increasingly being developed as world models. Accordingly, evaluating them in this role extends beyond the apparent appearance of generated videos to the inherent reactivity of the worlds they depict: the ability to infer from the scene state how the world should react and to generate plausible consequences not explicitly described in the input. Yet existing benchmarks mainly assess visual quality or explicit instruction fulfillment by checking whether requested actions and interaction outcomes are realized, leaving inherent reactivity underexamined. We introduce WorldExam, a hierarchical diagnostic benchmark spanning four levels: Visual Quality, Control Adherence, Spatial Consistency, and World Reactivity. It comprises 1,474 cases across eight dedicated tasks and supports unified evaluation of camera-, action-, and language-driven model paradigms. The World Reactivity level evaluates scene-conditioned reactions and goal-directed behaviors beyond what is explicitly specified in the input. Evaluation of 20 representative models reveals a clear capability split. Camera-driven models excel at camera control, but their interfaces do not support dynamic interaction; action-driven models control subjects more precisely but often leave the world unresponsive; and language-driven models perform better on interaction but follow complex controls less faithfully. No model combines broad task coverage with consistently strong performance, showing that high visual quality and explicit instruction fulfillment do not guarantee inherent reactivity.

View arXiv page View PDF Project page GitHub 6 Add to collection

Community

Yuppie1204

Paper submitter about 4 hours ago

WorldExam is a hierarchical diagnostic benchmark for controllable video world models. Beyond visual appearance and explicit instruction following, it asks whether a model preserves a coherent world and exhibits inherent reactivity. Across camera-, action-, and language-driven interfaces, WorldExam organizes 1,474 test cases into four diagnostic levels and eight dedicated tasks under a unified evaluation protocol.
Project Website: https://worldexam.github.io/