{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.01057", "kind": "paper", "url": "https://arxiv.org/abs/2606.01057"}, {"id": "gh:gaoypeng/3dcodebench", "kind": "repo", "url": "https://github.com/gaoypeng/3dcodebench"}, {"kind": "website", "url": "https://www.3dcodebench.com/"}], "authors": ["Yipeng Gao", "Lei Shu", "Genzhi Ye", "Xi Xiong", "Ameesh Makadia", "Meiqi Guo", "Laurent Itti", "Jindong Chen"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.01057", "source_url": "https://arxiv.org/abs/2606.01057"}, "categories": ["General AI", "Geometric reasoning", "Multimodal"], "description": {"en": "3DCodeBench evaluates vision-language model agents on procedural 3D modeling by converting text and image references into Blender Python code. It includes 212 object categories with ground-truth scripts, and scores outputs on executability, image similarity (SigLIP-2/DINOv3), 3D shape distance (Chamfer/Uni3D), and LLM-as-judge, plus a human-preference ranking platform."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.01057", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.01057", "title": "3DCodeBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of Southern California", "Google (United States)"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.01057", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.01057", "source_url": "https://openalex.org/W7163169950"}, "key": "claire-radar:2606.01057", "modality": null, "name": "3DCodeBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_3dcodebench_14ee1d5d", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "ef759b1301da55dc6ea9209d6b07b075b0b462cd2e6ad31f23a439ab5dfe00e3", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.01057"}, "publisher": null, "released": "2026-05-31", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.01057", "source_url": "https://arxiv.org/abs/2606.01057"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-01057", "source": "claire_radar", "source_benchmark_id": "2606.01057", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Multimodal", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/gaoypeng/3dcodebench", "githubScope": "benchmark_repo", "githubStars": 98, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 5, "hfPaperUrl": "https://huggingface.co/papers/2606.01057", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "fresh"}, "capabilities": ["Geometric reasoning"], "confidence": "High", "construction": "Transform Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "e46eeef8ddb34438972594cea22941243dba369ff169288b03aa926d38021f6a", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "3DCodeBench evaluates vision-language model agents on procedural 3D modeling by converting text and image references into Blender Python code. It includes 212 object categories with ground-truth scripts, and scores outputs on executability, image similarity (SigLIP-2/DINOv3), 3D shape distance (Chamfer/Uni3D), and LLM-as-judge, plus a human-preference ranking platform.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "In this paper, we propose 3DCodeBench, a systematic benchmark for evaluating vision-language model (VLM) agents for procedural 3D generation in 3D modeling software."}, "familyId": "bmf_151d2c52aca3", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/gaoypeng/3dcodebench", "url": "https://github.com/gaoypeng/3dcodebench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/gaoypeng/3dcodebench", "url": "https://huggingface.co/papers/2606.01057"}, {"basis": "huggingface-paper-association", "linkedUrl": "https://github.com/gaoypeng/3dcodebench", "url": "https://huggingface.co/api/papers/2606.01057"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/gaoypeng/3dcodebench/blob/main/CONTRIBUTING.md", "url": "https://www.3dcodebench.com/"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/gaoypeng/3dcodebench/issues/new", "url": "https://www.3dcodebench.com/"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/gaoypeng/3dcodebench", "url": "https://www.3dcodebench.com/"}], "scope": "benchmark_repo", "url": "https://github.com/gaoypeng/3dcodebench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_3dcodebench_14ee1d5d", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/gaoypeng/3dcodebench", "data": null, "hfPaper": "https://huggingface.co/papers/2606.01057", "paper": "https://arxiv.org/abs/2606.01057", "pdf": "https://arxiv.org/pdf/2606.01057", "project": "https://www.3dcodebench.com/", "report": "https://arxiv.org/abs/2606.01057"}, "metrics": [], "motivation": "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.", "name": "3DCodeBench", "oneLine": "3DCodeBench evaluates vision-language model agents on procedural 3D modeling by converting text and image references into Blender Python code. It includes 212 object categories with ground-truth scripts, and scores outputs on executability, image similarity (SigLIP-2/DINOv3), 3D shape distance (Chamfer/Uni3D), and LLM-as-judge, plus a human-preference ranking platform.", "paperTitle": "3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.01057", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Google", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/gaoypeng/3dcodebench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-31", "source": {"authors": ["Yipeng Gao", "Lei Shu", "Genzhi Ye", "Xi Xiong", "Ameesh Makadia", "Meiqi Guo", "Laurent Itti", "Jindong Chen"], "categories": ["cs.CV", "cs.AI", "cs.GR", "cs.LG"], "id": "2606.01057", "title": "3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code", "type": "arxiv", "url": "https://arxiv.org/abs/2606.01057"}, "sourceUpdatedAt": "2026-05-31T06:59:49Z", "topics": ["Multimodal"], "venueAttempts": [], "whyItMatters": "There is no standardized way to compare model abilities in procedural 3D code generation. This benchmark provides a fixed protocol and dataset, enabling reproducible evaluation and comparison across models and coding-agent settings, useful for selecting models or guiding development of procedural modeling capabilities."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
