{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.10803", "kind": "paper", "url": "https://arxiv.org/abs/2606.10803"}, {"id": "gh:modalitydance/phystool-bench", "kind": "repo", "url": "https://github.com/ModalityDance/PhysTool-Bench"}], "authors": ["Zhixin Ma", "Yutong Zhou", "Yongqi Li", "Chong-Wah Ngo", "Wenjie Li"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.10803", "source_url": "https://arxiv.org/abs/2606.10803"}, "categories": ["Agents", "Agents & Tool Use", "General AI", "Tool use"], "description": {"en": "PhysTool-Bench evaluates multimodal LLMs on physical tool use through two tasks: recognizing all tools in a scene and selecting and sequencing tools for a given task, using 2,510 queries over 2,678 real-world tools."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.10803", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.10803", "title": "PhysTool-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Singapore Management University", "Hong Kong Polytechnic University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.10803", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.10803", "source_url": "https://openalex.org/W7164183780"}, "key": "claire-radar:2606.10803", "modality": null, "name": "PhysTool-Bench", "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_phystool-bench_e6d3d9ac", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "e5809c40967bb6baf1d9a36bc0b511bb9773004f5e3d5f3abd507429c99a2aea", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.10803"}, "publisher": null, "released": "2026-06-09", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.10803", "source_url": "https://arxiv.org/abs/2606.10803"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-10803", "source": "claire_radar", "source_benchmark_id": "2606.10803", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Agents & Tool Use", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/ModalityDance/PhysTool-Bench", "githubScope": "benchmark_repo", "githubStars": 0, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 3, "hfPaperUrl": "https://huggingface.co/papers/2606.10803", "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": "2026-09-24", "state": "stale"}}, "status": "fresh"}, "capabilities": ["Tool use"], "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": "a6a97b05aa973e34994b7af031a29d2daf8f7f3edeaaf56c4d8b92a3b7adf350", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "PhysTool-Bench evaluates multimodal LLMs on physical tool use through two tasks: recognizing all tools in a scene and selecting and sequencing tools for a given task, using 2,510 queries over 2,678 real-world tools.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "To address this gap, we introduce PhysTool-Bench, the first physical tool-use benchmark designed to evaluate MLLMs' ability to comprehend real-world scenarios, identify physical tools, and plan their use."}, "familyId": "bmf_2635f961d28a", "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/ModalityDance/PhysTool-Bench", "url": "https://github.com/ModalityDance/PhysTool-Bench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/ModalityDance/PhysTool-Bench", "url": "https://huggingface.co/papers/2606.10803"}], "scope": "benchmark_repo", "url": "https://github.com/ModalityDance/PhysTool-Bench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_phystool-bench_e6d3d9ac", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/ModalityDance/PhysTool-Bench", "data": null, "hfPaper": "https://huggingface.co/papers/2606.10803", "paper": "https://arxiv.org/abs/2606.10803", "pdf": "https://arxiv.org/pdf/2606.10803", "project": null, "report": "https://arxiv.org/abs/2606.10803"}, "metrics": [], "motivation": "Multimodal Large Language Models (MLLMs) excel at utilizing digital APIs and increasingly serve as the \"brain\" of embodied AI, instructing robots to interact with the physical world.", "name": "PhysTool-Bench", "oneLine": "PhysTool-Bench evaluates multimodal LLMs on physical tool use through two tasks: recognizing all tools in a scene and selecting and sequencing tools for a given task, using 2,510 queries over 2,678 real-world tools.", "paperTitle": "Beyond APIs: Probing the Limits of MLLMs in Physical Tool Use", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.10803", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "ModalityDance", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/ModalityDance/PhysTool-Bench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-09", "source": {"authors": ["Zhixin Ma", "Yutong Zhou", "Yongqi Li", "Chong-Wah Ngo", "Wenjie Li"], "categories": ["cs.CL", "cs.AI", "cs.CV"], "id": "2606.10803", "title": "Beyond APIs: Probing the Limits of MLLMs in Physical Tool Use", "type": "arxiv", "url": "https://arxiv.org/abs/2606.10803"}, "sourceUpdatedAt": "2026-06-09T12:49:11Z", "topics": ["Agents"], "venueAttempts": [], "whyItMatters": "Physical tool use is underexplored in MLLMs; this benchmark isolates recognition and planning deficits, supporting progress in embodied AI and practical human-robot collaboration."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
