{"record": {"aliases": ["SWE-Refactor-Bench"], "artifacts": [{"kind": "paper", "url": "https://arxiv.org/abs/2608.23564"}], "audits": [], "categories": ["coding_agent"], "description": {"en": "20 whole-repository migrations only, so one task is 5 points of the mean score; the composite score gates on Migration Audit and all-or-nothing Behavioural Tests before Agentic Verification adds partial credit, and the harness and effort level are part of the result, so scores from different harnesses or effort levels are not the same measurement."}, "documents": [{"document_type": "benchmark_leaderboard", "id": "model_reports:swe_refactor_bench_results_table", "model_name": null, "organization": "Einsia.AI", "published": "2026-08-24", "retrieved_at": "2026-11-05", "source": "model_reports", "source_id": "swe_refactor_bench_results_table", "source_url": "https://arxiv.org/abs/2608.23564", "title": "SWE Refactor Bench results table"}], "evidence_summary": {"document_count": 1, "model_count": 8, "model_count_basis": "source_model_id"}, "key": "model-reports:swe_refactor_bench", "modality": null, "name": "SWE Refactor Bench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"registry": "data/model_cards.yml", "score_archive": "data/benchmark_scores.yml", "source_url": "https://arxiv.org/abs/2608.23564"}, "publisher": null, "released": "2026-08-24", "released_reference": {"basis": "benchmark_release", "source_key": "model-reports:swe_refactor_bench", "source_url": "https://arxiv.org/abs/2608.23564"}, "schema_version": 1, "sizes": [], "slug": "swe_refactor_bench", "source": "model_reports", "source_benchmark_id": "swe_refactor_bench"}, "schema_version": 1, "scores_by_source": {"model_reports": {"rows": [{"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: high\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Claude Opus 5", "model_id": "[\"Anthropic\",\"Claude Opus 5\"]", "model_name": "Claude Opus 5", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: high\u0000Claude Opus 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: high\u0000Claude Opus 5", "organization": "Anthropic", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: high", "rank_in_source_response": null, "raw_value": "34.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 34.5, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: high\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Claude Sonnet 5", "model_id": "[\"Anthropic\",\"Claude Sonnet 5\"]", "model_name": "Claude Sonnet 5", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: high\u0000Claude Sonnet 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: high\u0000Claude Sonnet 5", "organization": "Anthropic", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: high", "rank_in_source_response": null, "raw_value": "6.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 6.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: low\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Claude Opus 5", "model_id": "[\"Anthropic\",\"Claude Opus 5\"]", "model_name": "Claude Opus 5", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: low\u0000Claude Opus 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: low\u0000Claude Opus 5", "organization": "Anthropic", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: low", "rank_in_source_response": null, "raw_value": "20.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 20.5, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: low\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Claude Sonnet 5", "model_id": "[\"Anthropic\",\"Claude Sonnet 5\"]", "model_name": "Claude Sonnet 5", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: low\u0000Claude Sonnet 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: low\u0000Claude Sonnet 5", "organization": "Anthropic", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: low", "rank_in_source_response": null, "raw_value": "4.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 4.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Claude Opus 5", "model_id": "[\"Anthropic\",\"Claude Opus 5\"]", "model_name": "Claude Opus 5", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000Claude Opus 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000Claude Opus 5", "organization": "Anthropic", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max", "rank_in_source_response": null, "raw_value": "31.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 31.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Claude Sonnet 5", "model_id": "[\"Anthropic\",\"Claude Sonnet 5\"]", "model_name": "Claude Sonnet 5", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000Claude Sonnet 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000Claude Sonnet 5", "organization": "Anthropic", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max", "rank_in_source_response": null, "raw_value": "8.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 8.5, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "DeepSeek V4 Flash", "model_id": "[\"DeepSeek\",\"DeepSeek V4 Flash\"]", "model_name": "DeepSeek V4 Flash", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000DeepSeek V4 Flash", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000DeepSeek V4 Flash", "organization": "DeepSeek", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max", "rank_in_source_response": null, "raw_value": "7.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 7.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GLM 5.2", "model_id": "[\"Z.ai\",\"GLM 5.2\"]", "model_name": "GLM 5.2", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000GLM 5.2", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000GLM 5.2", "organization": "Z.ai", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max", "rank_in_source_response": null, "raw_value": "6.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 6.5, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Kimi K3", "model_id": "[\"Moonshot AI\",\"Kimi K3\"]", "model_name": "Kimi K3", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000Kimi K3", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000Kimi K3", "organization": "Moonshot AI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max", "rank_in_source_response": null, "raw_value": "19.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 19.5, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Qwen 3.8 Max", "model_id": "[\"Qwen\",\"Qwen 3.8 Max\"]", "model_name": "Qwen 3.8 Max", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000Qwen 3.8 Max", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max\u0000Qwen 3.8 Max", "organization": "Qwen", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: max", "rank_in_source_response": null, "raw_value": "10.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 10.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: medium\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Claude Opus 5", "model_id": "[\"Anthropic\",\"Claude Opus 5\"]", "model_name": "Claude Opus 5", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: medium\u0000Claude Opus 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: medium\u0000Claude Opus 5", "organization": "Anthropic", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: medium", "rank_in_source_response": null, "raw_value": "28.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 28.5, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: medium\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Claude Sonnet 5", "model_id": "[\"Anthropic\",\"Claude Sonnet 5\"]", "model_name": "Claude Sonnet 5", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: medium\u0000Claude Sonnet 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: medium\u0000Claude Sonnet 5", "organization": "Anthropic", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: medium", "rank_in_source_response": null, "raw_value": "15.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 15.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: xhigh\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Claude Opus 5", "model_id": "[\"Anthropic\",\"Claude Opus 5\"]", "model_name": "Claude Opus 5", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: xhigh\u0000Claude Opus 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: xhigh\u0000Claude Opus 5", "organization": "Anthropic", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: xhigh", "rank_in_source_response": null, "raw_value": "47.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 47.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: xhigh\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "Claude Sonnet 5", "model_id": "[\"Anthropic\",\"Claude Sonnet 5\"]", "model_name": "Claude Sonnet 5", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: xhigh\u0000Claude Sonnet 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: xhigh\u0000Claude Sonnet 5", "organization": "Anthropic", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: xhigh", "rank_in_source_response": null, "raw_value": "9.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 9.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: high\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Luna", "model_id": "[\"OpenAI\",\"GPT-5.6 Luna\"]", "model_name": "GPT-5.6 Luna", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: high\u0000GPT-5.6 Luna", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: high\u0000GPT-5.6 Luna", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: high", "rank_in_source_response": null, "raw_value": "4.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 4.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: high\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Sol", "model_id": "[\"OpenAI\",\"GPT-5.6 Sol\"]", "model_name": "GPT-5.6 Sol", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: high\u0000GPT-5.6 Sol", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: high\u0000GPT-5.6 Sol", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: high", "rank_in_source_response": null, "raw_value": "19.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 19.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: low\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Luna", "model_id": "[\"OpenAI\",\"GPT-5.6 Luna\"]", "model_name": "GPT-5.6 Luna", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: low\u0000GPT-5.6 Luna", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: low\u0000GPT-5.6 Luna", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: low", "rank_in_source_response": null, "raw_value": "4.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 4.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: low\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Sol", "model_id": "[\"OpenAI\",\"GPT-5.6 Sol\"]", "model_name": "GPT-5.6 Sol", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: low\u0000GPT-5.6 Sol", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: low\u0000GPT-5.6 Sol", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: low", "rank_in_source_response": null, "raw_value": "7.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 7.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: max\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Luna", "model_id": "[\"OpenAI\",\"GPT-5.6 Luna\"]", "model_name": "GPT-5.6 Luna", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: max\u0000GPT-5.6 Luna", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: max\u0000GPT-5.6 Luna", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: max", "rank_in_source_response": null, "raw_value": "10.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 10.5, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: max\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Sol", "model_id": "[\"OpenAI\",\"GPT-5.6 Sol\"]", "model_name": "GPT-5.6 Sol", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: max\u0000GPT-5.6 Sol", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: max\u0000GPT-5.6 Sol", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: max", "rank_in_source_response": null, "raw_value": "28.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 28.5, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: medium\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Luna", "model_id": "[\"OpenAI\",\"GPT-5.6 Luna\"]", "model_name": "GPT-5.6 Luna", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: medium\u0000GPT-5.6 Luna", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: medium\u0000GPT-5.6 Luna", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: medium", "rank_in_source_response": null, "raw_value": "0.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 0.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: medium\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Sol", "model_id": "[\"OpenAI\",\"GPT-5.6 Sol\"]", "model_name": "GPT-5.6 Sol", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: medium\u0000GPT-5.6 Sol", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: medium\u0000GPT-5.6 Sol", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: medium", "rank_in_source_response": null, "raw_value": "6.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 6.5, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: none\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Luna", "model_id": "[\"OpenAI\",\"GPT-5.6 Luna\"]", "model_name": "GPT-5.6 Luna", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: none\u0000GPT-5.6 Luna", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: none\u0000GPT-5.6 Luna", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: none", "rank_in_source_response": null, "raw_value": "4.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 4.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: none\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Sol", "model_id": "[\"OpenAI\",\"GPT-5.6 Sol\"]", "model_name": "GPT-5.6 Sol", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: none\u0000GPT-5.6 Sol", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: none\u0000GPT-5.6 Sol", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: none", "rank_in_source_response": null, "raw_value": "4.0", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 4.0, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: xhigh\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Luna", "model_id": "[\"OpenAI\",\"GPT-5.6 Luna\"]", "model_name": "GPT-5.6 Luna", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: xhigh\u0000GPT-5.6 Luna", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: xhigh\u0000GPT-5.6 Luna", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: xhigh", "rank_in_source_response": null, "raw_value": "5.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 5.5, "value_kind": "number"}, {"benchmark_id": "swe_refactor_bench", "comparable_group": "[\"model-reports:swe_refactor_bench\",\"swe_refactor_bench_20_tasks\",\"20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: xhigh\"]", "date_precision": "document_publication", "document_id": "model_reports:swe_refactor_bench_results_table", "instrument": "swe_refactor_bench_20_tasks", "key": "model-reports:swe_refactor_bench", "measured_by": "Einsia.AI", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Sol", "model_id": "[\"OpenAI\",\"GPT-5.6 Sol\"]", "model_name": "GPT-5.6 Sol", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: xhigh\u0000GPT-5.6 Sol", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: xhigh\u0000GPT-5.6 Sol", "organization": "OpenAI", "protocol": "20 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Codex; effort: xhigh", "rank_in_source_response": null, "raw_value": "9.5", "read_from": "pdf_text", "reported_at": "2026-08-24", "reported_by": "third_party", "reported_date": "2026-08-24", "series_id": "model_reports:swe_refactor_bench:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "swe_refactor_bench_results_table", "source_title": "SWE Refactor Bench results table", "source_url": "https://arxiv.org/abs/2608.23564", "value": 9.5, "value_kind": "number"}], "series": {"declared_max": null, "direction": "higher_is_better", "direction_basis": "reported_metric", "display_scale": null, "first_score_record": {"date_precision": "document_publication", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: high\u0000Claude Opus 5", "reported_at": "2026-08-24", "source_url": "https://arxiv.org/abs/2608.23564"}, "first_score_report": {"date_precision": "document_publication", "obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: high\u0000Claude Opus 5", "reported_at": "2026-08-24", "source_url": "https://arxiv.org/abs/2608.23564"}, "key": "model-reports:swe_refactor_bench", "max_score_contradicted": false, "metric": "score", "observation_count": 26, "observed_max": 47.0, "score_summary": {"display_max": 47.0, "display_multiplier": 1, "model_count": 8, "model_count_basis": "source_model_id", "numeric_count": 26, "raw_max": 47.0, "raw_min": 0.0, "source_reference": {"obs_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: xhigh\u0000Claude Opus 5", "observation_id": "curated\u0000swe_refactor_bench\u0000swe_refactor_bench_results_table\u0000swe_refactor_bench_20_tasks\u000020 whole-repository migrations; score per equation (4), mean over runs, out of 100; harness: Claude Code; effort: xhigh\u0000Claude Opus 5", "reported_at": "2026-08-24", "reported_date": "2026-08-24", "source_id": "swe_refactor_bench_results_table", "source_url": "https://arxiv.org/abs/2608.23564"}, "unit": "percent"}, "series_id": "model_reports:swe_refactor_bench:default", "unit": "percent"}}}, "siblings": [{"key": "claire-radar:2608.23564", "name": "SWE Refactor Bench", "relation": "equivalent", "slug": "claire-radar-2608-23564", "source": "claire_radar"}]}