{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.28480", "kind": "paper", "url": "https://arxiv.org/abs/2606.28480"}, {"id": "gh:facebookresearch/tua-bench", "kind": "repo", "url": "https://github.com/facebookresearch/TUA-Bench"}, {"kind": "website", "url": "https://www.tuabench.ai"}], "authors": ["Shoufa Chen", "Luyuan Wang", "Xuan Yang", "Zhiheng Liu", "Yuren Cong", "Yuanfeng Ji", "Feiyan Zhou", "Xiaohui Zhang", "Fanny Yang", "Belinda Zeng"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.28480", "source_url": "https://arxiv.org/abs/2606.28480"}, "categories": ["Code & Software", "General AI", "cs.SE"], "description": {"en": "Evaluates terminal-use agents on 120 real-world tasks across five families (document editing, email, web info, scientific/engineering workflows) using execution-based scoring."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.28480", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.28480", "title": "TUA-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Duke University", "Stanford University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.28480", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.28480", "source_url": "https://openalex.org/W7166665335"}, "key": "claire-radar:2606.28480", "modality": null, "name": "TUA-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_tua-bench_56e7e977", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "082c9900ecef468f3d591eaafd0f7adc8ddee8c717ee53791beb51138f7ea0b8", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.28480"}, "publisher": null, "released": "2026-06-26", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.28480", "source_url": "https://arxiv.org/abs/2606.28480"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-28480", "source": "claire_radar", "source_benchmark_id": "2606.28480", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Code & Software", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/facebookresearch/TUA-Bench", "githubScope": "benchmark_repo", "githubStars": 48, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 48, "hfPaperUrl": "https://huggingface.co/papers/2606.28480", "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": "6b4c90791bb32c4e309ff9509a785f3211b5bd10d48a36c6d542ace7f7d1455d", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Evaluates terminal-use agents on 120 real-world tasks across five families (document editing, email, web info, scientific/engineering workflows) using execution-based scoring.", "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 introduce TUA-Bench, a general-purpose benchmark for terminal-use agents."}, "familyId": "bmf_6458a26e1270", "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/facebookresearch/TUA-Bench", "url": "https://github.com/facebookresearch/TUA-Bench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/facebookresearch/TUA-Bench", "url": "https://huggingface.co/papers/2606.28480"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/facebookresearch/TUA-Bench", "url": "https://www.tuabench.ai"}], "scope": "benchmark_repo", "url": "https://github.com/facebookresearch/TUA-Bench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_tua-bench_56e7e977", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/facebookresearch/TUA-Bench", "data": null, "hfPaper": "https://huggingface.co/papers/2606.28480", "paper": "https://arxiv.org/abs/2606.28480", "pdf": "https://arxiv.org/pdf/2606.28480", "project": "https://www.tuabench.ai", "report": "https://arxiv.org/abs/2606.28480"}, "metrics": [], "motivation": "As large language models and harness frameworks continue to advance, agents operating in terminals are increasingly capable of performing a broader range of general computer-use tasks beyond coding.", "name": "TUA-Bench", "oneLine": "Evaluates terminal-use agents on 120 real-world tasks across five families (document editing, email, web info, scientific/engineering workflows) using execution-based scoring.", "paperTitle": "TUA-Bench: A Benchmark for General-Purpose Terminal-Use Agents", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.28480", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Facebook Research", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/facebookresearch/TUA-Bench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-26", "source": {"authors": ["Shoufa Chen", "Luyuan Wang", "Xuan Yang", "Zhiheng Liu", "Yuren Cong", "Yuanfeng Ji", "Feiyan Zhou", "Xiaohui Zhang", "Fanny Yang", "Belinda Zeng"], "categories": ["cs.SE", "cs.AI"], "id": "2606.28480", "title": "TUA-Bench: A Benchmark for General-Purpose Terminal-Use Agents", "type": "arxiv", "url": "https://arxiv.org/abs/2606.28480"}, "sourceUpdatedAt": "2026-06-26T17:59:51Z", "topics": ["cs.SE"], "venueAttempts": [], "whyItMatters": "Provides a broad, realistic terminal benchmark beyond coding, showing frontier agents achieve only 65.8% success, highlighting gaps in general-purpose digital work."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
