3DCodeBench is a systematic benchmark and human evaluation platform for assessing vision-language model agents in generating procedural 3D assets via code in 3D modeling software.</p>\n","updatedAt":"2026-06-02T04:16:18.128Z","author":{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","fullname":"taesiri","name":"taesiri","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":309,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8059126138687134},"editors":["taesiri"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.01057","authors":[{"_id":"6a1e58ed808ddbc3c7d43de6","name":"Yipeng Gao","hidden":false},{"_id":"6a1e58ed808ddbc3c7d43de7","name":"Lei Shu","hidden":false},{"_id":"6a1e58ed808ddbc3c7d43de8","name":"Genzhi Ye","hidden":false},{"_id":"6a1e58ed808ddbc3c7d43de9","name":"Xi Xiong","hidden":false},{"_id":"6a1e58ed808ddbc3c7d43dea","name":"Ameesh Makadia","hidden":false},{"_id":"6a1e58ed808ddbc3c7d43deb","name":"Meiqi Guo","hidden":false},{"_id":"6a1e58ed808ddbc3c7d43dec","name":"Laurent Itti","hidden":false},{"_id":"6a1e58ed808ddbc3c7d43ded","name":"Jindong Chen","hidden":false}],"publishedAt":"2026-05-31T00:00:00.000Z","submittedOnDailyAt":"2026-06-02T00:00:00.000Z","title":"3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code","submittedOnDailyBy":{"_id":"6039478ab3ecf716b1a5fd4d","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/6039478ab3ecf716b1a5fd4d/_Thy4E7taiSYBLKxEKJbT.jpeg","isPro":true,"fullname":"taesiri","user":"taesiri","type":"user","name":"taesiri"},"summary":"Procedural 3D modeling through code is emerging as a versatile paradigm, offering deterministic, engine-ready, and precisely editable assets that neural 3D generators inherently lack. Authoring such procedural content, however, demands deep expertise in 3D software APIs, parametric design, and code-level geometric reasoning. In this paper, we propose 3DCodeBench, a systematic benchmark for evaluating vision-language model (VLM) agents for procedural 3D generation in 3D modeling software. Specifically, 3DCodeBench evaluates how effectively 12 advanced VLMs can serve as procedural 3D modelers by translating text and image references into procedural code for 3D modeling software. Recognizing that automated metrics may not fully capture the perceptual quality of 3D shapes, we build 3DCodeArena, a ranking platform based on pairwise human preferences over generated 3D outputs. From extensive evaluations and results, we observe that: (1) Failures mostly arise from API mismatches, while successful renders still suffer from disconnected or floating 3D geometric components. (2) Test-time scaling, such as higher thinking budgets and multi-turn refinement, improves performance overall. Our findings highlight a critical need for high-quality procedural coding data to advance commercial VLMs. Furthermore, effective procedural 3D modeling requires a robust execution environment that provides high-fidelity feedback for iterative refinement. We release 3DCodeBench, including the curated large-scale dataset of multimodal (text/image) prompts, procedural code, 3D object triplets, evaluation protocol, and the public 3DCodeArena platform as a foundational toolkit for exploring VLM-based procedural 3D modelers.","upvotes":3,"discussionId":"6a1e58ed808ddbc3c7d43dee","projectPage":"https://www.3dcodebench.com/","ai_summary":"Vision-language models are evaluated for procedural 3D modeling tasks through a benchmark and ranking platform that assess their ability to translate text and images into executable 3D code.","ai_keywords":["vision-language model","procedural 3D generation","3D modeling software","multimodal prompts","procedural code","3D object triplets","test-time scaling","iterative refinement"],"organization":{"_id":"60f6cbb2852126bac698c89e","name":"deepmind","fullname":"Deepmind","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1638956859875-5f1158120c833276f61f1a84.jpeg"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"69cca9c9eb9cdf88f2a0127b","avatarUrl":"/avatars/f79daeb0eb10b9210c8fa4524700b376.svg","isPro":false,"fullname":"Лебедев Наталья","user":"BENJAMINW60","type":"user"},{"_id":"67312401d433c6b122c38202","avatarUrl":"/avatars/599523d1412ab01671aa7f86c14d508a.svg","isPro":false,"fullname":"Yipeng Gao","user":"YipengGao","type":"user"},{"_id":"69ccaff8712fc1e42730a3ae","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/PwSbFzsl-ZCqc9CJmPuu-.png","isPro":false,"fullname":"林葵","user":"mateomartin30","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"60f6cbb2852126bac698c89e","name":"deepmind","fullname":"Deepmind","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/1638956859875-5f1158120c833276f61f1a84.jpeg"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.01057.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2606.01057 in a model README.md to link it from this page.

Cite arxiv.org/abs/2606.01057 in a Space README.md to link it from this page.

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