{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2607.05155", "kind": "paper", "url": "https://arxiv.org/abs/2607.05155"}, {"id": "gh:bytedance-seed/edgebench", "kind": "repo", "url": "https://github.com/ByteDance-Seed/EdgeBench"}, {"id": "hf:bytedance-seed/edgebench", "kind": "dataset", "url": "https://huggingface.co/datasets/ByteDance-Seed/EdgeBench"}, {"kind": "website", "url": "https://edge-bench.org/"}], "authors": ["Deyao Zhu", "Xin Zhou", "Shengling Qin", "Xuekai Zhu", "Hangliang Ding", "Shu Zhong", "Zixin Wen", "Zhonglin Xie", "Chenhui Gou", "Linxuan Ren", "Yueyang Wang", "Junfeng Zhong", "Rui Liu", "Tian Gao", "Yangguang Lin", "Jingyuan Zhang", "Maojia Song", "Xuan Qi", "Jinhong Wu", "Chenyang Zhang", "Yinzhu Piao", "Ziru Niu", "Hongbin Lin", "Lingxiang Meng", "Peng Tang", "Chengyao Tang", "Shanyu Wu", "Huanyu Zheng", "Yu Liu", "Liya Zhu", "He Wang", "Ming Ding", "Ziyu Wan", "Hao Liu", "Sibo Wang", "Haotian Zhu", "Xintian Zhang", "Nan Chai", "Yipeng Liu", "Panhao Lai", "Sihang Yuan", "Zixin Su", "Ge Zhang", "Wangchunshu Zhou", "Yantao Du", "Wenhao Huang", "Guang Shi"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.05155", "source_url": "https://arxiv.org/abs/2607.05155"}, "categories": ["Agents", "Agents & Tool Use", "Environment Learning", "Environment learning", "General AI", "Learning from feedback", "Long-horizon task execution", "Mathematics & Formal Science", "Scaling Laws", "Scientific Research & AI for Science", "Software & AI Compute"], "description": {"en": "EdgeBench evaluates autonomous agents on 134 real-world tasks across scientific discovery, software engineering, optimization, knowledge work, formal mathematics, and games. Each task requires 12+ hours of continuous interaction with multi-level feedback. Scoring tracks agent performance over time (at 2,4,6,8,10,12 hours). A public leaderboard is maintained; 51 tasks and evaluation framework are open-sourced."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.05155", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.05155", "title": "EdgeBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.05155", "modality": null, "name": "EdgeBench", "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-reviewed", "display_eligible": "unknown", "origin_record_id": "bm_edgebench_88dc3c92", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "96be3ed73446a6eb2abb60729face50581edc9dc1ceb60e9618a26a5509a4ce5", "relation": "introduces", "review_state": "source-reviewed", "reviewed_at": "2026-08-19", "source_url": "https://arxiv.org/abs/2607.05155"}, "publication_dates": [{"basis": "paper_first_version", "date": "2026-07-06", "source_url": "https://arxiv.org/abs/2607.05155"}], "publisher": null, "released": "2026-07-02", "released_reference": {"basis": "first_public", "source_key": "claire-radar:2607.05155", "source_url": "https://edge-bench.org/"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-05155", "source": "claire_radar", "source_benchmark_id": "2607.05155", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI", "Scientific Research & AI for Science", "Software & AI Compute", "Mathematics & Formal Science"], "area": "Agents & Tool Use", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/ByteDance-Seed/EdgeBench", "githubScope": "benchmark_repo", "githubStars": 459, "hfDailySubmittedAt": null, "hfDataset": "ByteDance-Seed/EdgeBench", "hfDatasetDownloads": 4592, "hfDatasetLikes": 84, "hfPaperUpvotes": 15, "hfPaperUrl": "https://huggingface.co/papers/2607.05155", "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": "2026-10-04", "state": "stale"}, "hfDatasetLikes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-04", "state": "stale"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-04", "state": "stale"}}, "status": "fresh"}, "capabilities": ["Environment learning", "Long-horizon task execution", "Learning from feedback"], "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": "bf947e2af808593cdcd52d4e4578c40a22f3628308d6650745e5a65db4ffa6b3", "model": "deepseek-v4-flash"}, "curation": {"reviewedAt": "2026-08-19", "sources": ["https://arxiv.org/abs/2607.05155", "https://edge-bench.org/", "https://github.com/ByteDance-Seed/EdgeBench", "https://huggingface.co/datasets/ByteDance-Seed/EdgeBench"], "state": "source-reviewed"}, "dataStatus": "primary-source-reviewed", "demo": false, "description": "EdgeBench evaluates autonomous agents on 134 real-world tasks across scientific discovery, software engineering, optimization, knowledge work, formal mathematics, and games. Each task requires 12+ hours of continuous interaction with multi-level feedback. Scoring tracks agent performance over time (at 2,4,6,8,10,12 hours). A public leaderboard is maintained; 51 tasks and evaluation framework are open-sourced.", "domainCuration": {"method": "primary-source-review-v1", "state": "reviewed"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "named benchmark identity and task release stated in nearby sentences", "evaluation protocol evidence"], "snippet": "This discovery stems from EdgeBench, a suite of 134 real world tasks with ultra-long horizons, spanning scientific discovery, software engineering, combinatorial optimization, professional knowledge work, formal mathematics, and interactive games. […] We publicly release 51 tasks and our full evaluation framework to accelerate the study of how agents learn from real world experience."}, "familyId": "bmf_2c4d59d4e34e", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 1, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/ByteDance-Seed/EdgeBench", "url": "https://github.com/ByteDance-Seed/EdgeBench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/ByteDance-Seed/EdgeBench", "url": "https://huggingface.co/papers/2607.05155"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/ByteDance-Seed/EdgeBench", "url": "https://edge-bench.org/"}, {"basis": "linked-from-data", "linkedUrl": "https://github.com/ByteDance-Seed/EdgeBench/blob/main/assets/wechat_qr.jpg", "url": "https://huggingface.co/datasets/ByteDance-Seed/EdgeBench"}, {"basis": "linked-from-data", "linkedUrl": "https://github.com/ByteDance-Seed/EdgeBench", "url": "https://huggingface.co/datasets/ByteDance-Seed/EdgeBench"}], "scope": "benchmark_repo", "url": "https://github.com/ByteDance-Seed/EdgeBench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_edgebench_88dc3c92", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/ByteDance-Seed/EdgeBench", "data": "https://huggingface.co/datasets/ByteDance-Seed/EdgeBench", "hfPaper": "https://huggingface.co/papers/2607.05155", "paper": "https://arxiv.org/abs/2607.05155", "pdf": "https://arxiv.org/pdf/2607.05155", "project": "https://edge-bench.org/", "report": "https://arxiv.org/abs/2607.05155"}, "metrics": [], "motivation": "Pretraining scaling laws reveal that model capability improves predictably with data and compute.", "name": "EdgeBench", "oneLine": "EdgeBench evaluates autonomous agents on 134 real-world tasks across scientific discovery, software engineering, optimization, knowledge work, formal mathematics, and games. Each task requires 12+ hours of continuous interaction with multi-level feedback. Scoring tracks agent performance over time (at 2,4,6,8,10,12 hours). A public leaderboard is maintained; 51 tasks and evaluation framework are open-sourced.", "paperTitle": "EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.05155", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "ByteDance Seed", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/ByteDance-Seed/EdgeBench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releaseDates": {"firstPublicAt": "2026-07-02", "paperV1At": "2026-07-06"}, "releasedAt": "2026-07-06", "source": {"authors": ["Deyao Zhu", "Xin Zhou", "Shengling Qin", "Xuekai Zhu", "Hangliang Ding", "Shu Zhong", "Zixin Wen", "Zhonglin Xie", "Chenhui Gou", "Linxuan Ren", "Yueyang Wang", "Junfeng Zhong", "Rui Liu", "Tian Gao", "Yangguang Lin", "Jingyuan Zhang", "Maojia Song", "Xuan Qi", "Jinhong Wu", "Chenyang Zhang", "Yinzhu Piao", "Ziru Niu", "Hongbin Lin", "Lingxiang Meng", "Peng Tang", "Chengyao Tang", "Shanyu Wu", "Huanyu Zheng", "Yu Liu", "Liya Zhu", "He Wang", "Ming Ding", "Ziyu Wan", "Hao Liu", "Sibo Wang", "Haotian Zhu", "Xintian Zhang", "Nan Chai", "Yipeng Liu", "Panhao Lai", "Sihang Yuan", "Zixin Su", "Ge Zhang", "Wangchunshu Zhou", "Yantao Du", "Wenhao Huang", "Guang Shi"], "categories": ["cs.CL", "cs.LG"], "id": "2607.05155", "title": "EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments", "type": "arxiv", "url": "https://arxiv.org/abs/2607.05155"}, "sourceUpdatedAt": "2026-07-06T14:39:22Z", "topics": ["Agents", "Environment Learning", "Scaling Laws"], "venueAttempts": [], "whyItMatters": "EdgeBench fills a gap in evaluating agents' ability to learn from real-world environments over extended periods, providing a standardized protocol for comparing long-horizon learning capabilities. It offers a public leaderboard and open-source tasks, enabling reproducible comparisons and tracking of agent improvement over time, which is valuable for model selection and development."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
