{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.13148", "kind": "paper", "url": "https://arxiv.org/abs/2606.13148"}], "authors": ["Dat Tien Nguyen", "Thao Nguyen", "Fadillah Adamsyah Maani", "Huy M. Le", "Muhammad Umer Sheikh", "Numan Saeed", "Muhammad Haris Khan", "Salman Khan"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.13148", "source_url": "https://arxiv.org/abs/2606.13148"}, "categories": ["General AI", "Information retrieval", "Language & Knowledge", "Planning", "Reasoning"], "description": {"en": "TerraBench is a benchmark for grounded Earth-science reasoning, built on TerraAgent, a ReAct-style framework that couples LLM planning with scientific tools. It includes 403 tasks across three tracks and eight domains with 24,500 verified steps."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.13148", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.13148", "title": "TerraBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Mohamed bin Zayed University of Artificial Intelligence"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.13148", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.13148", "source_url": "https://openalex.org/W7164571725"}, "key": "claire-radar:2606.13148", "modality": null, "name": "TerraBench", "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_terrabench_679be7e3", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "6f5c18634010ed4646d28c81a6e34db0c2b2d022f2e72d417a480998110431e3", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.13148"}, "publisher": null, "released": "2026-06-11", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.13148", "source_url": "https://arxiv.org/abs/2606.13148"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-13148", "source": "claire_radar", "source_benchmark_id": "2606.13148", "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": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2606.13148", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-01", "state": "stale"}}, "status": "stale"}, "capabilities": ["Reasoning", "Planning", "Information retrieval"], "confidence": "High", "construction": "Interactive Environment", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "cf0c3dd97e1cf343dfa8a50d16094860427a3ca3115107919789d9da73088f21", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "TerraBench is a benchmark for grounded Earth-science reasoning, built on TerraAgent, a ReAct-style framework that couples LLM planning with scientific tools. It includes 403 tasks across three tracks and eight domains with 24,500 verified steps.", "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": "We introduce TerraBench, a benchmark for grounded Earth-science reasoning, built on TerraAgent, a ReAct-style executable framework that interleaves reasoning, tool calls, and observations to couple LLM planning with scientific tools for environmental retrieval, geospatial processing, simulation, and artifact-backed computation."}, "familyId": "bmf_b1de429d892f", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_terrabench_679be7e3", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.13148", "paper": "https://arxiv.org/abs/2606.13148", "pdf": "https://arxiv.org/pdf/2606.13148", "project": null, "report": "https://arxiv.org/abs/2606.13148"}, "metrics": [], "motivation": "Climate and environmental decision-making increasingly requires reasoning across heterogeneous inputs, including gridded physical data, satellite imagery, geospatial context, and simulator outputs.", "name": "TerraBench", "oneLine": "TerraBench is a benchmark for grounded Earth-science reasoning, built on TerraAgent, a ReAct-style framework that couples LLM planning with scientific tools. It includes 403 tasks across three tracks and eight domains with 24,500 verified steps.", "paperTitle": "TerraBench: Can Agents Reason Over Heterogeneous Earth-System Data?", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.13148", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-11", "source": {"authors": ["Dat Tien Nguyen", "Thao Nguyen", "Fadillah Adamsyah Maani", "Huy M. Le", "Muhammad Umer Sheikh", "Numan Saeed", "Muhammad Haris Khan", "Salman Khan"], "categories": ["cs.AI"], "id": "2606.13148", "title": "TerraBench: Can Agents Reason Over Heterogeneous Earth-System Data?", "type": "arxiv", "url": "https://arxiv.org/abs/2606.13148"}, "sourceUpdatedAt": "2026-07-01T17:31:23Z", "topics": ["Reasoning"], "venueAttempts": [], "whyItMatters": "Earth-science workflows require reasoning over heterogeneous data types. TerraBench unifies these capabilities in a single interface, with process-level metrics and tolerance-aware scoring."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
