{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2608.09873", "kind": "paper", "url": "https://arxiv.org/abs/2608.09873"}, {"id": "gh:sci-vbench/sci-vbench", "kind": "repo", "url": "https://github.com/sci-vbench/sci-vbench"}], "authors": ["Diandian Zhang", "Tingyu Song", "Lin Fu", "Zheyuan Yang", "Yilun Zhao"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.09873", "source_url": "https://arxiv.org/abs/2608.09873"}, "categories": ["Factuality", "General AI", "Reasoning", "Vision & 3D"], "description": {"en": "Sci-VBench evaluates text-to-video generation across 1,253 expert-annotated examples in 60 scientific subjects. It requires temporally rich videos demonstrating scientific reasoning and knowledge-grounded synthesis. Scoring covers four dimensions: prompt grounding, scientific correctness, spatiotemporal consistency, and low-level perceptual fidelity, using a rubric-based protocol with MLLM judges."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.09873", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.09873", "title": "Sci-VBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Zhejiang University", "University of Chinese Academy of Sciences", "Tongji University", "Yale University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2608.09873", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2608.09873", "source_url": "https://openalex.org/W7202181426"}, "key": "claire-radar:2608.09873", "modality": null, "name": "Sci-VBench", "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_sci-vbench_8bc62acf", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "38710c62a838710f34142764452f7e0f73a9ed1a3f6f625caaffcc6807cc8416", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2608.09873"}, "publisher": null, "released": "2026-08-10", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.09873", "source_url": "https://arxiv.org/abs/2608.09873"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-09873", "source": "claire_radar", "source_benchmark_id": "2608.09873", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/sci-vbench/sci-vbench", "githubScope": "benchmark_repo", "githubStars": 4, "hfDailySubmittedAt": "2026-08-11T00:00:00.000Z", "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 28, "hfPaperUrl": "https://huggingface.co/papers/2608.09873", "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": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}}, "status": "fresh"}, "capabilities": ["Reasoning", "Factuality"], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "0bea81b4ff8609c221e7040b2848b4fe21dbaa5d03563ab582c5f638e731ee58", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Sci-VBench evaluates text-to-video generation across 1,253 expert-annotated examples in 60 scientific subjects. It requires temporally rich videos demonstrating scientific reasoning and knowledge-grounded synthesis. Scoring covers four dimensions: prompt grounding, scientific correctness, spatiotemporal consistency, and low-level perceptual fidelity, using a rubric-based protocol with MLLM judges.", "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"], "snippet": "We introduce Sci-VBench, a comprehensive benchmark for evaluating knowledge- and reasoning-intensive video generation across scientific domains."}, "familyId": "bmf_063001d7adb8", "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/sci-vbench/sci-vbench", "url": "https://github.com/sci-vbench/sci-vbench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/sci-vbench/sci-vbench", "url": "https://huggingface.co/papers/2608.09873"}], "scope": "benchmark_repo", "url": "https://github.com/sci-vbench/sci-vbench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_sci-vbench_8bc62acf", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/sci-vbench/sci-vbench", "data": null, "hfPaper": "https://huggingface.co/papers/2608.09873", "paper": "https://arxiv.org/abs/2608.09873", "pdf": "https://arxiv.org/pdf/2608.09873", "project": null, "report": "https://arxiv.org/abs/2608.09873"}, "metrics": [], "motivation": "We introduce Sci-VBench, a comprehensive benchmark for evaluating knowledge- and reasoning-intensive video generation across scientific domains.", "name": "Sci-VBench", "oneLine": "Sci-VBench evaluates text-to-video generation across 1,253 expert-annotated examples in 60 scientific subjects. It requires temporally rich videos demonstrating scientific reasoning and knowledge-grounded synthesis. Scoring covers four dimensions: prompt grounding, scientific correctness, spatiotemporal consistency, and low-level perceptual fidelity, using a rubric-based protocol with MLLM judges.", "paperTitle": "Sci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.09873", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {"90d": {"confidence": "Medium", "coverage": 0.7, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.21483102985913266, "value": 4}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.5949830216993645, "value": 28}}, "confidence": "Medium", "coverage": 0.7, "rank": 175, "score": 36}, "method": "attention-ranking-v13", "rank": 175, "score": 36}}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-08-10", "source": {"authors": ["Diandian Zhang", "Tingyu Song", "Lin Fu", "Zheyuan Yang", "Yilun Zhao"], "categories": ["cs.CV", "cs.AI"], "id": "2608.09873", "title": "Sci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains", "type": "arxiv", "url": "https://arxiv.org/abs/2608.09873"}, "sourceUpdatedAt": "2026-08-10T17:27:32Z", "topics": ["Reasoning"], "venueAttempts": [], "whyItMatters": "Existing video generation benchmarks focus on surface realism, leaving scientific and causal correctness unmeasured. Sci-VBench provides a public protocol and open dataset to compare models on knowledge-intensive generation, revealing gaps between visual quality and reliable scientific dynamics."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
