{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.22557", "kind": "paper", "url": "https://arxiv.org/abs/2606.22557"}, {"id": "gh:jetastra/macagentbench", "kind": "repo", "url": "https://github.com/JetAstra/MacAgentBench"}], "authors": ["Yikun Fu", "Bowen Fu", "Zhenyu Wu", "Shuang Cheng", "Xiaowei Sun", "Bowen Yang", "Zehao Li", "Yibo Zhao", "Zichen Ding", "Zhoumianze Liu", "Shijie Wang", "Biqing Qi", "Bowen Zhou"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.22557", "source_url": "https://arxiv.org/abs/2606.22557"}, "categories": ["Agents", "General AI", "Language & Knowledge"], "description": {"en": "MacAgentBench evaluates computer use agents on macOS desktop automation, with 676 tasks across 25 applications, including GUI and CLI interactions. It uses deterministic rule-based evaluation and fine-grained multi-checkpoint scoring to assess sub-goal completion."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.22557", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.22557", "title": "MacAgentBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.22557", "modality": null, "name": "MacAgentBench", "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_macagentbench_782bfad8", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "b4ec115f7c5a1e357dcda9c9784ac953c9db8bfabd64107d7effe3ea46edd54f", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.22557"}, "publisher": null, "released": "2026-06-21", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.22557", "source_url": "https://arxiv.org/abs/2606.22557"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-22557", "source": "claire_radar", "source_benchmark_id": "2606.22557", "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/JetAstra/MacAgentBench", "githubScope": "benchmark_repo", "githubStars": 52, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2606.22557", "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": "2026-10-04", "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": "826975e9c350fe0c6796ec56a396a5d639dfbc794ec5598e314bad1d4492384e", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "MacAgentBench evaluates computer use agents on macOS desktop automation, with 676 tasks across 25 applications, including GUI and CLI interactions. It uses deterministic rule-based evaluation and fine-grained multi-checkpoint scoring to assess sub-goal completion.", "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": "We present MacAgentBench, a comprehensive macOS agent benchmark comprising 676 tasks across 25 applications, with nearly 60% involving both GUI and CLI interaction."}, "familyId": "bmf_bd6b325f56e2", "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/JetAstra/MacAgentBench", "url": "https://github.com/JetAstra/MacAgentBench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/JetAstra/MacAgentBench", "url": "https://huggingface.co/papers/2606.22557"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/JetAstra/MacAgentBench.", "url": "https://arxiv.org/abs/2606.22557"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/JetAstra/MacAgentBench", "url": "https://arxiv.org/abs/2606.22557"}], "scope": "benchmark_repo", "url": "https://github.com/JetAstra/MacAgentBench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_macagentbench_782bfad8", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/JetAstra/MacAgentBench", "data": null, "hfPaper": "https://huggingface.co/papers/2606.22557", "paper": "https://arxiv.org/abs/2606.22557", "pdf": "https://arxiv.org/pdf/2606.22557", "project": null, "report": "https://arxiv.org/abs/2606.22557"}, "metrics": [], "motivation": "Computer use agents (CUAs) have advanced rapidly in desktop automation, and a growing number of users deploy CUAs such as OpenClaw on Mac Mini for always-on automation.", "name": "MacAgentBench", "oneLine": "MacAgentBench evaluates computer use agents on macOS desktop automation, with 676 tasks across 25 applications, including GUI and CLI interactions. It uses deterministic rule-based evaluation and fine-grained multi-checkpoint scoring to assess sub-goal completion.", "paperTitle": "MacAgentBench: Benchmarking AI Agents on Real-World macOS Desktop", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.22557", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "JetAstra", "organizationType": "community", "role": "benchmark-publisher", "sourceUrl": "https://github.com/JetAstra/MacAgentBench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-21", "source": {"authors": ["Yikun Fu", "Bowen Fu", "Zhenyu Wu", "Shuang Cheng", "Xiaowei Sun", "Bowen Yang", "Zehao Li", "Yibo Zhao", "Zichen Ding", "Zhoumianze Liu", "Shijie Wang", "Biqing Qi", "Bowen Zhou"], "categories": ["cs.AI", "cs.CL", "cs.HC"], "id": "2606.22557", "title": "MacAgentBench: Benchmarking AI Agents on Real-World macOS Desktop", "type": "arxiv", "url": "https://arxiv.org/abs/2606.22557"}, "sourceUpdatedAt": "2026-06-21T15:32:57Z", "topics": ["Agents"], "venueAttempts": [], "whyItMatters": "MacAgentBench addresses the need for benchmarks that capture framework-augmented agent capabilities and partial progress on long-horizon, multi-application tasks, providing a more granular comparison for real-world desktop automation."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
