{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.30573", "kind": "paper", "url": "https://arxiv.org/abs/2606.30573"}, {"id": "gh:scaleapi/swe-interact", "kind": "repo", "url": "https://github.com/scaleapi/SWE-Interact"}], "authors": ["Mohit Raghavendra", "Anisha Gunjal", "Aakash Sabharwal", "Yunzhong He"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.30573", "source_url": "https://arxiv.org/abs/2606.30573"}, "categories": ["Agents & Tool Use", "Code", "Software & AI Compute"], "description": {"en": "SWE-Interact evaluates coding agents on multi-turn, interactive software engineering tasks where a simulated user provides vague instructions, reveals requirements progressively, and gives feedback. The benchmark comprises 75 tasks and measures agents' ability to discover user intent, adapt to evolving requirements, and build on prior work."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.30573", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.30573", "title": "SWE-INTERACT"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.30573", "modality": null, "name": "SWE-INTERACT", "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-interact_046bbdbf", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "0fcb412232d6647277478afd46ae9656631956097a4afdbe4aa07f17f48e067e", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.30573"}, "publisher": null, "released": "2026-06-29", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.30573", "source_url": "https://arxiv.org/abs/2606.30573"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-30573", "source": "claire_radar", "source_benchmark_id": "2606.30573", "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/scaleapi/SWE-Interact", "githubScope": "benchmark_repo", "githubStars": 28, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 7, "hfPaperUrl": "https://huggingface.co/papers/2606.30573", "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": "f9fc79903ecb54517711cf6a755aafb2701ab8e2f8bdce223851471f86fae8cb", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "SWE-Interact evaluates coding agents on multi-turn, interactive software engineering tasks where a simulated user provides vague instructions, reveals requirements progressively, and gives feedback. The benchmark comprises 75 tasks and measures agents' ability to discover user intent, adapt to evolving requirements, and build on prior work.", "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-Interact, a new testbed for evaluating coding agents on multi-turn, interactive, user-driven software engineering tasks."}, "familyId": "bmf_93acd7a96f92", "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/scaleapi/SWE-Interact", "url": "https://github.com/scaleapi/SWE-Interact"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/scaleapi/SWE-Interact", "url": "https://huggingface.co/papers/2606.30573"}], "scope": "benchmark_repo", "url": "https://github.com/scaleapi/SWE-Interact", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_swe-interact_046bbdbf", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Software & Cloud"], "links": {"code": "https://github.com/scaleapi/SWE-Interact", "data": null, "hfPaper": "https://huggingface.co/papers/2606.30573", "paper": "https://arxiv.org/abs/2606.30573", "pdf": "https://arxiv.org/pdf/2606.30573", "project": null, "report": "https://arxiv.org/abs/2606.30573"}, "metrics": [], "motivation": "We introduce SWE-Interact, a new testbed for evaluating coding agents on multi-turn, interactive, user-driven software engineering tasks.", "name": "SWE-INTERACT", "oneLine": "SWE-Interact evaluates coding agents on multi-turn, interactive software engineering tasks where a simulated user provides vague instructions, reveals requirements progressively, and gives feedback. The benchmark comprises 75 tasks and measures agents' ability to discover user intent, adapt to evolving requirements, and build on prior work.", "paperTitle": "SWE-INTERACT: Reimagining SWE Benchmarks as User-Driven Long-Horizon Coding Sessions", "primaryDomain": "Software & AI Compute", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.30573", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Scale AI", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/scaleapi/SWE-Interact"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-29", "source": {"authors": ["Mohit Raghavendra", "Anisha Gunjal", "Aakash Sabharwal", "Yunzhong He"], "categories": ["cs.LG"], "id": "2606.30573", "title": "SWE-INTERACT: Reimagining SWE Benchmarks as User-Driven Long-Horizon Coding Sessions", "type": "arxiv", "url": "https://arxiv.org/abs/2606.30573"}, "sourceUpdatedAt": "2026-06-29T17:17:45Z", "topics": ["Code"], "venueAttempts": [], "whyItMatters": "Existing SWE benchmarks focus on single-turn autonomous implementation, but real developer workflows are interactive. SWE-Interact fills the gap by measuring performance on long-horizon, user-driven tasks, showing that strong single-turn performance does not reliably transfer. This provides a more realistic evaluation axis for coding agents and guides development of models that can collaborate effectively with users."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
