{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2607.07946", "kind": "paper", "url": "https://arxiv.org/abs/2607.07946"}, {"id": "gh:datacurve-ai/deep-swe", "kind": "repo", "url": "https://github.com/datacurve-ai/deep-swe"}, {"id": "hf:datacurve/deep-swe", "kind": "dataset", "url": "https://huggingface.co/datasets/datacurve/deep-swe"}, {"kind": "website", "url": "https://deepswe.datacurve.ai/"}], "authors": ["Wenqi Huang", "Charley Lee", "Leonard Tng", "Serena Ge"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.07946", "source_url": "https://arxiv.org/abs/2607.07946"}, "categories": ["Code & Software", "Software & AI Compute", "cs.SE"], "description": {"en": "Evaluates coding agents on 113 original, long-horizon software engineering tasks across 91 open-source repositories in five languages. Tasks are written from scratch, with hand-written verifiers that check requested functionality and accept any correct implementation."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.07946", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.07946", "title": "DeepSWE"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.07946", "modality": null, "name": "DeepSWE", "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_deepswe_ebdcd38e", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "60600fea77d0ff014f795138dee52c0d591303a5634ece6cbfff09cb2efb274e", "relation": "introduces", "review_state": "source-reviewed", "reviewed_at": "2026-08-19", "source_url": "https://arxiv.org/abs/2607.07946"}, "publication_dates": [{"basis": "paper_first_version", "date": "2026-07-08", "source_url": "https://arxiv.org/abs/2607.07946"}], "publisher": null, "released": "2026-05-18", "released_reference": {"basis": "first_public", "source_key": "claire-radar:2607.07946", "source_url": "https://datacurve.ai/research"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-07946", "source": "claire_radar", "source_benchmark_id": "2607.07946", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Software & AI Compute"], "area": "Code & Software", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/datacurve-ai/deep-swe", "githubScope": "benchmark_repo", "githubStars": 1792, "hfDailySubmittedAt": null, "hfDataset": "datacurve/deep-swe", "hfDatasetDownloads": 1149, "hfDatasetLikes": 83, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2607.07946", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetLikes": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfPaperUpvotes": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}}, "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": "e890c3ad2a987f7362993def072c47c833df4d1ceec4c5eb76afd49724188295", "model": "deepseek-v4-flash"}, "curation": {"reviewedAt": "2026-08-19", "sources": ["https://datacurve.ai/research", "https://deepswe.datacurve.ai/", "https://github.com/datacurve-ai/deep-swe", "https://arxiv.org/abs/2607.07946"], "state": "source-reviewed"}, "dataStatus": "primary-source-reviewed", "demo": false, "description": "Evaluates coding agents on 113 original, long-horizon software engineering tasks across 91 open-source repositories in five languages. Tasks are written from scratch, with hand-written verifiers that check requested functionality and accept any correct implementation.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We release the benchmark, its verifiers, and the full record of evaluation trajectories."}, "familyId": "bmf_3f3a44e7f6ef", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 93, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/datacurve-ai/deep-swe", "url": "https://github.com/datacurve-ai/deep-swe"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/datacurve-ai/deep-swe", "url": "https://huggingface.co/papers/2607.07946"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/datacurve-ai/deep-swe", "url": "https://deepswe.datacurve.ai/"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/datacurve-ai/deep-swe", "url": "https://arxiv.org/abs/2607.07946"}, {"basis": "linked-from-data", "linkedUrl": "https://github.com/datacurve-ai/deep-swe", "url": "https://huggingface.co/datasets/datacurve/deep-swe"}], "scope": "benchmark_repo", "url": "https://github.com/datacurve-ai/deep-swe", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_deepswe_ebdcd38e", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Software & Cloud"], "links": {"code": "https://github.com/datacurve-ai/deep-swe", "data": "https://huggingface.co/datasets/datacurve/deep-swe", "hfPaper": "https://huggingface.co/papers/2607.07946", "paper": "https://arxiv.org/abs/2607.07946", "pdf": "https://arxiv.org/pdf/2607.07946", "project": "https://deepswe.datacurve.ai/", "report": "https://arxiv.org/abs/2607.07946"}, "metrics": [], "motivation": "DeepSWE is a benchmark of 113 original, long-horizon software engineering tasks for evaluating coding agents.", "name": "DeepSWE", "oneLine": "Evaluates coding agents on 113 original, long-horizon software engineering tasks across 91 open-source repositories in five languages. Tasks are written from scratch, with hand-written verifiers that check requested functionality and accept any correct implementation.", "paperTitle": "DeepSWE: Measuring Frontier Coding Agents on Original, Long-Horizon Engineering Tasks", "primaryDomain": "Software & AI Compute", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.07946", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "DataCurve", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://deepswe.datacurve.ai/"}], "ranking": {"90d": {"confidence": "High", "coverage": 1.0, "datasetDownloadRank": 16, "datasetRankPopulation": 85, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 1.0, "value": 1792}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": 0.7257527799751998, "value": 1149}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.0, "value": 0}}, "confidence": "High", "coverage": 1.0, "rank": 2, "score": 77}, "method": "attention-ranking-v13", "rank": 2, "score": 77}}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releaseDates": {"firstPublicAt": "2026-05-18", "paperV1At": "2026-07-08"}, "releasedAt": "2026-07-08", "source": {"authors": ["Wenqi Huang", "Charley Lee", "Leonard Tng", "and Serena Ge"], "categories": ["cs.SE", "cs.LG"], "id": "2607.07946", "title": "DeepSWE: Measuring Frontier Coding Agents on Original, Long-Horizon Engineering Tasks", "type": "arxiv", "url": "https://arxiv.org/abs/2607.07946"}, "sourceUpdatedAt": "2026-07-08T21:45:34Z", "topics": ["cs.SE"], "venueAttempts": [], "whyItMatters": "Addresses the gap of benchmarks relying on mined fixes and inherited tests, which can overstate model capability due to pretraining exposure and rigid grading. Provides a reusable evaluation path with verifiers and trajectories for assessing genuine problem-solving ability."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
