{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.18943", "kind": "paper", "url": "https://arxiv.org/abs/2606.18943"}, {"id": "gh:google-deepmind/physics-iq-benchmark", "kind": "repo", "url": "https://github.com/google-deepmind/physics-iq-benchmark"}], "authors": ["Tim Rädsch", "Yuki M Asano", "Hilde Kuehne", "Stefan Bauer", "Priyank Jaini", "Robert Geirhos", "Carsten T. Lüth"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.18943", "source_url": "https://arxiv.org/abs/2606.18943"}, "categories": ["General AI", "Vision & 3D", "cs.CV"], "description": {"en": "A revised benchmark for measuring whether video generative models reproduce the physical behavior observed in real-world experiment videos."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.18943", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.18943", "title": "Physics-IQ Verified"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.18943", "modality": null, "name": "Physics-IQ Verified", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_physics-iq-verified_dda3be6d", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "8abcbee6a80583671221526b79a93bed4acdb388440210cc0ad3785cab4911bc", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T17:16:21.828872Z", "source_url": "https://arxiv.org/abs/2606.18943"}, "publisher": null, "released": "2026-06-17", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.18943", "source_url": "https://arxiv.org/abs/2606.18943"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-18943", "source": "claire_radar", "source_benchmark_id": "2606.18943", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "Its audit of an existing video-physics benchmark and reported model-ranking changes may generate near-term interest among video-model evaluators.", "score": 6}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T17:16:21.828872Z", "inputHash": "f5dbcbb537bebdcfcad979b40e4a9ac18c8bb1554e3140f6ece067efb227f29f", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "Physics-IQ Verified", "canonicalNameSource": "paper_title", "decisionReason": "The source formally names the benchmark, identifies its evaluation task, states a comparable sample-level scoring rule, and supplies a public benchmark-code path.", "evaluationEvidence": {"reuse": "The code for the benchmark can be accessed at https://github.com/google-deepmind/physics-iq-benchmark", "scoring": "a sample-level scoring system that weights each sample and metric equally", "task": "quantifies this explicitly by comparing model-generated videos to real-world videos of physical experiments"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T17:16:21.828872Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "A revised benchmark for measuring whether video generative models reproduce the physical behavior observed in real-world experiment videos.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "In this work, we present a systematic audit of the Physics-IQ benchmark, expose shortcomings and propose three solutions that sharpen how we can measure physical understanding of VGMs."}, "familyId": "bmf_1e2378605fad", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_physics-iq-verified_dda3be6d", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/google-deepmind/physics-iq-benchmark", "data": null, "paper": "https://arxiv.org/abs/2606.18943", "pdf": "https://arxiv.org/pdf/2606.18943", "project": null, "report": "https://arxiv.org/abs/2606.18943"}, "metrics": [], "motivation": "Video generative models ( VGMs) have become a new frontier that can be used not just for video generation but for a multitude of downstream tasks, including world modeling.", "name": "Physics-IQ Verified", "oneLine": "A revised benchmark for measuring whether video generative models reproduce the physical behavior observed in real-world experiment videos.", "paperTitle": "Physics-IQ Verified", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.18943", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "publishers": [{"name": "Google DeepMind", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/google-deepmind/physics-iq-benchmark"}], "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-06-17", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Tim R\\\"adsch", "Yuki M Asano", "Hilde Kuehne", "Stefan Bauer", "Priyank Jaini", "Robert Geirhos", "Carsten T. L\\\"uth"], "categories": ["cs.CV"], "id": "2606.18943", "title": "Physics-IQ Verified", "type": "arxiv", "url": "https://arxiv.org/abs/2606.18943"}, "sourceUpdatedAt": "2026-06-17T11:23:52Z", "topics": ["cs.CV"], "venueAttempts": [], "whyItMatters": "It audits and refines an existing physical-understanding evaluation while defining equal weighting across samples and metrics, supporting more reliable model comparisons."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
