{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.29957", "kind": "paper", "url": "https://arxiv.org/abs/2606.29957"}, {"id": "gh:togetherbench/swe-together", "kind": "repo", "url": "https://github.com/Togetherbench/SWE-Together"}], "authors": ["Yifan Wu", "Zhuokai Zhao", "Songlin Li", "Ho Hin Lee", "Jiacheng Zhu", "Shirley Wu", "Tianhe Yu", "Serena Li", "Lizhu Zhang", "Xiangjun Fan", "Shengzhi Li"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.29957", "source_url": "https://arxiv.org/abs/2606.29957"}, "categories": ["Agents", "Agents & Tool Use", "Software & AI Compute"], "description": {"en": "SWE-Together evaluates coding agents in multi-turn interactive user sessions reconstructed from real user-agent interactions. It comprises 109 repository-level tasks with a reactive LLM-based user simulator, measuring final repository correctness and the number of corrective feedback turns."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.29957", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.29957", "title": "SWE-Together"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.29957", "modality": null, "name": "SWE-Together", "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_swe-together_260e2683", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "b5eebac6ede04d3c47b0c8f9ab9323cdbbce6a0d4f399a0757a09fd7835ef967", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.29957"}, "publisher": null, "released": "2026-06-29", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.29957", "source_url": "https://arxiv.org/abs/2606.29957"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-29957", "source": "claire_radar", "source_benchmark_id": "2606.29957", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Software & AI Compute"], "area": "Agents & Tool Use", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/Togetherbench/SWE-Together", "githubScope": "benchmark_repo", "githubStars": 70, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 13, "hfPaperUrl": "https://huggingface.co/papers/2606.29957", "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": "Interactive Environment", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "a2b28d10dee2abdbcde94ad261c709e00c47c052a270c697ee14dce0e07a2dc2", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "SWE-Together evaluates coding agents in multi-turn interactive user sessions reconstructed from real user-agent interactions. It comprises 109 repository-level tasks with a reactive LLM-based user simulator, measuring final repository correctness and the number of corrective feedback turns.", "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"], "snippet": "We introduce SWE-Together, a multi-turn benchmark reconstructed from real user-agent coding sessions."}, "familyId": "bmf_123ecb3dd733", "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/Togetherbench/SWE-Together", "url": "https://github.com/Togetherbench/SWE-Together"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/Togetherbench/SWE-Together", "url": "https://huggingface.co/papers/2606.29957"}], "scope": "benchmark_repo", "url": "https://github.com/Togetherbench/SWE-Together", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_swe-together_260e2683", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Software & Cloud"], "links": {"code": "https://github.com/Togetherbench/SWE-Together", "data": null, "hfPaper": "https://huggingface.co/papers/2606.29957", "paper": "https://arxiv.org/abs/2606.29957", "pdf": "https://arxiv.org/pdf/2606.29957", "project": null, "report": "https://arxiv.org/abs/2606.29957"}, "metrics": [], "motivation": "Most coding-agent benchmarks are static: an agent receives a complete task description up front and is judged only by its final code.", "name": "SWE-Together", "oneLine": "SWE-Together evaluates coding agents in multi-turn interactive user sessions reconstructed from real user-agent interactions. It comprises 109 repository-level tasks with a reactive LLM-based user simulator, measuring final repository correctness and the number of corrective feedback turns.", "paperTitle": "SWE-Together: Evaluating Coding Agents in Interactive User Sessions", "primaryDomain": "Software & AI Compute", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.29957", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Togetherbench", "organizationType": "community", "role": "benchmark-publisher", "sourceUrl": "https://github.com/Togetherbench/SWE-Together"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-29", "source": {"authors": ["Yifan Wu", "Zhuokai Zhao", "Songlin Li", "Ho Hin Lee", "Jiacheng Zhu", "Shirley Wu", "Tianhe Yu", "Serena Li", "Lizhu Zhang", "Xiangjun Fan", "Shengzhi Li"], "categories": ["cs.SE", "cs.AI"], "id": "2606.29957", "title": "SWE-Together: Evaluating Coding Agents in Interactive User Sessions", "type": "arxiv", "url": "https://arxiv.org/abs/2606.29957"}, "sourceUpdatedAt": "2026-06-29T08:35:15Z", "topics": ["Agents"], "venueAttempts": [], "whyItMatters": "Existing coding-agent benchmarks often evaluate static, single-turn tasks, missing the interactive nature of real coding assistance. SWE-Together provides a reproducible protocol for assessing agents as collaborators, capturing both task success and user effort, offering practical value for comparing agents in realistic settings."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
