{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.09426", "kind": "paper", "url": "https://arxiv.org/abs/2606.09426"}, {"id": "gh:weavebench/weavebench", "kind": "repo", "url": "https://github.com/weavebench/WeaveBench"}, {"id": "hf:wanlilll/weavebench", "kind": "dataset", "url": "https://huggingface.co/datasets/wanlilll/WeaveBench"}, {"kind": "website", "url": "https://weavebench.github.io/"}], "authors": ["Wanli Li", "Bowen Zhou", "Yunyao Yu", "Zhou Xu", "Yifan Yang", "Dongsheng Li", "Caihua Shan"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.09426", "source_url": "https://arxiv.org/abs/2606.09426"}, "categories": ["General AI", "Language & Knowledge", "cs.AI"], "description": {"en": "WeaveBench evaluates computer-use agents on 114 long-horizon tasks across 8 real-world work domains. Each task interleaves GUI interaction with command-line and code operations in a single trajectory, with scoring based on a trajectory-aware judge that detects fabricated evidence."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.09426", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.09426", "title": "WeaveBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.09426", "modality": null, "name": "WeaveBench", "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_weavebench_27cb863c", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "99481039175be5ed2a674c089235eeb88237998303565847c32d5915786a1f17", "relation": "introduces", "review_state": "source-reviewed", "reviewed_at": "2026-08-30", "source_url": "https://arxiv.org/abs/2606.09426"}, "publisher": null, "released": "2026-06-08", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.09426", "source_url": "https://arxiv.org/abs/2606.09426"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-09426", "source": "claire_radar", "source_benchmark_id": "2606.09426", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/weavebench/WeaveBench", "githubScope": "benchmark_repo", "githubStars": 169, "hfDailySubmittedAt": null, "hfDataset": "wanlilll/WeaveBench", "hfDatasetDownloads": 2203, "hfDatasetLikes": 8, "hfPaperUpvotes": 48, "hfPaperUrl": "https://huggingface.co/papers/2606.09426", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-25", "state": "stale"}, "hfDatasetLikes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-25", "state": "stale"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-24", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "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": "af8ec21e146fe5a6035ccd9fe211bba62061a42025bd35f8560232cc962ab2ae", "model": "deepseek-v4-flash"}, "curation": {"reviewedAt": "2026-08-30", "sources": ["https://arxiv.org/abs/2606.09426", "https://weavebench.github.io/", "https://github.com/weavebench/WeaveBench", "https://huggingface.co/datasets/wanlilll/WeaveBench"], "state": "source-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "WeaveBench evaluates computer-use agents on 114 long-horizon tasks across 8 real-world work domains. Each task interleaves GUI interaction with command-line and code operations in a single trajectory, with scoring based on a trajectory-aware judge that detects fabricated evidence.", "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": "Thus, we introduce WeaveBench, a long-horizon hybrid-interface benchmark with 114 tasks across 8 real-world work domains, grounded in real user requests and publicly verifiable artifacts."}, "familyId": "bmf_88fdd16b4aa1", "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/weavebench/WeaveBench", "url": "https://github.com/weavebench/WeaveBench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/weavebench/WeaveBench", "url": "https://huggingface.co/papers/2606.09426"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/weavebench/WeaveBench", "url": "https://weavebench.github.io/"}], "scope": "benchmark_repo", "url": "https://github.com/weavebench/WeaveBench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_weavebench_27cb863c", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/weavebench/WeaveBench", "data": "https://huggingface.co/datasets/wanlilll/WeaveBench", "hfPaper": "https://huggingface.co/papers/2606.09426", "paper": "https://arxiv.org/abs/2606.09426", "pdf": "https://arxiv.org/pdf/2606.09426", "project": "https://weavebench.github.io/", "report": "https://arxiv.org/abs/2606.09426"}, "metrics": [], "motivation": "Computer-use agents (CUAs) increasingly operate in runtimes that combine visual desktop control, command-line execution, code editing, browsers, and external tools.", "name": "WeaveBench", "oneLine": "WeaveBench evaluates computer-use agents on 114 long-horizon tasks across 8 real-world work domains. Each task interleaves GUI interaction with command-line and code operations in a single trajectory, with scoring based on a trajectory-aware judge that detects fabricated evidence.", "paperTitle": "WeaveBench: A Long-Horizon, Real-World Benchmark for Computer-Use Agents with Hybrid Interfaces", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.09426", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "WeaveBench team", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/weavebench/WeaveBench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-08", "source": {"authors": ["Wanli Li", "Bowen Zhou", "Yunyao Yu", "Zhou Xu", "Yifan Yang", "Dongsheng Li", "and Caihua Shan"], "categories": ["cs.AI"], "id": "2606.09426", "title": "WeaveBench: A Long-Horizon, Real-World Benchmark for Computer-Use Agents with Hybrid Interfaces", "type": "arxiv", "url": "https://arxiv.org/abs/2606.09426"}, "sourceUpdatedAt": "2026-07-06T14:25:02Z", "topics": ["cs.AI"], "venueAttempts": [], "whyItMatters": "Existing benchmarks often evaluate interfaces separately, leaving hybrid orchestration under-tested. WeaveBench fills this gap by requiring agents to combine GUI and CLI/code in realistic tasks, and exposes that outcome-only grading overestimates performance."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
