{"record": {"aliases": ["LifeSci Diamond", "LifeSci-Diamond"], "artifacts": [{"kind": "website", "url": "https://www.arca.inc/blog/lifesci-diamond"}], "audits": [], "categories": ["health"], "description": {"en": "55 life-science research tasks; 10 public, 45 held out. Strict pass requires every rubric line fully met and no penalty fires. Scores are publisher-reported by Arca, which publishes the benchmark and fields evaluated systems, so they are not independent measurements. Sample scores are not comparable to the full benchmark. Only 44 of 55 answer keys are specialist-written or reviewed; 11 rest on primary sources without expert review, and 6 November additions are not second-reviewed."}, "documents": [{"document_type": "benchmark_leaderboard", "id": "model_reports:lifesci_diamond_arca_blog", "model_name": null, "organization": "Arca", "published": null, "retrieved_at": "2026-11-05", "source": "model_reports", "source_id": "lifesci_diamond_arca_blog", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "title": "LifeSci Diamond results (Arca blog)"}], "evidence_summary": {"document_count": 1, "model_count": 22, "model_count_basis": "source_model_id"}, "key": "model-reports:lifesci_diamond", "modality": null, "name": "LifeSci Diamond", "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://www.arca.inc/blog/lifesci-diamond"}, "publisher": null, "released": null, "released_reference": null, "schema_version": 1, "sizes": [], "slug": "lifesci_diamond", "source": "model_reports", "source_benchmark_id": "lifesci_diamond"}, "schema_version": 1, "scores_by_source": {"model_reports": {"rows": [{"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Arca Auto", "model_id": "[\"Arca\",\"Arca Auto\"]", "model_name": "Arca Auto", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca Auto", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca Auto", "organization": "Arca", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task", "rank_in_source_response": null, "raw_value": "20.0", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 20.0, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Arca with Claude Opus 5", "model_id": "[\"Arca\",\"Arca with Claude Opus 5\"]", "model_name": "Arca with Claude Opus 5", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca with Claude Opus 5", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca with Claude Opus 5", "organization": "Arca", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task", "rank_in_source_response": null, "raw_value": "9.1", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 9.1, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Arca with Claude Opus 5.5", "model_id": "[\"Arca\",\"Arca with Claude Opus 5.5\"]", "model_name": "Arca with Claude Opus 5.5", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca with Claude Opus 5.5", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca with Claude Opus 5.5", "organization": "Arca", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task", "rank_in_source_response": null, "raw_value": "12.7", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 12.7, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Arca with GPT-6 Astra", "model_id": "[\"Arca\",\"Arca with GPT-6 Astra\"]", "model_name": "Arca with GPT-6 Astra", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca with GPT-6 Astra", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca with GPT-6 Astra", "organization": "Arca", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task", "rank_in_source_response": null, "raw_value": "29.1", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 29.1, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Claude Fable 5", "model_id": "[\"Anthropic\",\"Claude Fable 5\"]", "model_name": "Claude Fable 5", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Fable 5", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Fable 5", "organization": "Anthropic", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "3.6", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 3.6, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Claude Fable 5.1", "model_id": "[\"Anthropic\",\"Claude Fable 5.1\"]", "model_name": "Claude Fable 5.1", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Fable 5.1", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Fable 5.1", "organization": "Anthropic", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "1.8", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 1.8, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Claude Opus 4.8", "model_id": "[\"Anthropic\",\"Claude Opus 4.8\"]", "model_name": "Claude Opus 4.8", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Opus 4.8", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Opus 4.8", "organization": "Anthropic", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "0.0", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 0.0, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Claude Opus 5", "model_id": "[\"Anthropic\",\"Claude Opus 5\"]", "model_name": "Claude Opus 5", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Opus 5", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Opus 5", "organization": "Anthropic", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "1.8", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 1.8, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Claude Opus 5.5", "model_id": "[\"Anthropic\",\"Claude Opus 5.5\"]", "model_name": "Claude Opus 5.5", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Opus 5.5", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Opus 5.5", "organization": "Anthropic", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "3.6", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 3.6, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Claude Sonnet 5", "model_id": "[\"Anthropic\",\"Claude Sonnet 5\"]", "model_name": "Claude Sonnet 5", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Sonnet 5", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Claude Sonnet 5", "organization": "Anthropic", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "1.8", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 1.8, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "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\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000GPT-5.6 Sol", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000GPT-5.6 Sol", "organization": "OpenAI", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "5.5", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 5.5, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "GPT-5.6 Terra", "model_id": "[\"OpenAI\",\"GPT-5.6 Terra\"]", "model_name": "GPT-5.6 Terra", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000GPT-5.6 Terra", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000GPT-5.6 Terra", "organization": "OpenAI", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "1.8", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 1.8, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "GPT-6 Astra", "model_id": "[\"OpenAI\",\"GPT-6 Astra\"]", "model_name": "GPT-6 Astra", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000GPT-6 Astra", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000GPT-6 Astra", "organization": "OpenAI", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "20.0", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 20.0, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "GPT-6.1 Sol", "model_id": "[\"OpenAI\",\"GPT-6.1 Sol\"]", "model_name": "GPT-6.1 Sol", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000GPT-6.1 Sol", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000GPT-6.1 Sol", "organization": "OpenAI", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "14.5", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 14.5, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "GPT-6.1 Terra", "model_id": "[\"OpenAI\",\"GPT-6.1 Terra\"]", "model_name": "GPT-6.1 Terra", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000GPT-6.1 Terra", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000GPT-6.1 Terra", "organization": "OpenAI", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "9.1", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 9.1, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Gemini 3.1 Pro Preview", "model_id": "[\"Google\",\"Gemini 3.1 Pro Preview\"]", "model_name": "Gemini 3.1 Pro Preview", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Gemini 3.1 Pro Preview", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Gemini 3.1 Pro Preview", "organization": "Google", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "0.0", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 0.0, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Gemini 3.7 Flash", "model_id": "[\"Google\",\"Gemini 3.7 Flash\"]", "model_name": "Gemini 3.7 Flash", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Gemini 3.7 Flash", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Gemini 3.7 Flash", "organization": "Google", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "1.8", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 1.8, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Gemini 4 Argon", "model_id": "[\"Google\",\"Gemini 4 Argon\"]", "model_name": "Gemini 4 Argon", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Gemini 4 Argon", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Gemini 4 Argon", "organization": "Google", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "0.0", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 0.0, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Grok 4.6", "model_id": "[\"xAI\",\"Grok 4.6\"]", "model_name": "Grok 4.6", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Grok 4.6", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Grok 4.6", "organization": "xAI", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "3.6", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 3.6, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Kimi K3", "model_id": "[\"Moonshot AI\",\"Kimi K3\"]", "model_name": "Kimi K3", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Kimi K3", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Kimi K3", "organization": "Moonshot AI", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "1.8", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 1.8, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Muse Spark 1.2", "model_id": "[\"Meta\",\"Muse Spark 1.2\"]", "model_name": "Muse Spark 1.2", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Muse Spark 1.2", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Muse Spark 1.2", "organization": "Meta", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "0.0", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 0.0, "value_kind": "number"}, {"benchmark_id": "lifesci_diamond", "comparable_group": "[\"model-reports:lifesci_diamond\",\"lifesci_diamond_full_55\",\"55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\"]", "date_precision": "document_publication", "document_id": "model_reports:lifesci_diamond_arca_blog", "instrument": "lifesci_diamond_full_55", "key": "model-reports:lifesci_diamond", "measured_by": "Arca", "measurement_kind": "benchmark_publisher_run", "model": "Muse Wonder", "model_id": "[\"Meta\",\"Muse Wonder\"]", "model_name": "Muse Wonder", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Muse Wonder", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded\u0000Muse Wonder", "organization": "Meta", "protocol": "55 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); plain API via Vercel AI Gateway reference harness, web_search tool (Exa); up to 40 model steps per turn; one answer per task; API IDs not pinned to model snapshots; temperature, top-p, seed, reasoning effort and max output tokens left to provider defaults, effective values not recorded", "rank_in_source_response": null, "raw_value": "9.1", "read_from": "html_text", "reported_at": "2026-11-04", "reported_by": "third_party", "reported_date": "2026-11-04", "series_id": "model_reports:lifesci_diamond:default", "source": "model_reports", "source_document_type": "benchmark_leaderboard", "source_id": "lifesci_diamond_arca_blog", "source_title": "LifeSci Diamond results (Arca blog)", "source_url": "https://www.arca.inc/blog/lifesci-diamond", "value": 9.1, "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\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca Auto", "reported_at": "2026-11-04", "source_url": "https://www.arca.inc/blog/lifesci-diamond"}, "first_score_report": {"date_precision": "document_publication", "obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca Auto", "reported_at": "2026-11-04", "source_url": "https://www.arca.inc/blog/lifesci-diamond"}, "key": "model-reports:lifesci_diamond", "max_score_contradicted": false, "metric": "strict_pass_rate", "observation_count": 22, "observed_max": 29.1, "score_summary": {"display_max": 29.1, "display_multiplier": 1, "model_count": 22, "model_count_basis": "source_model_id", "numeric_count": 22, "raw_max": 29.1, "raw_min": 0.0, "source_reference": {"obs_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca with GPT-6 Astra", "observation_id": "curated\u0000lifesci_diamond\u0000lifesci_diamond_arca_blog\u0000lifesci_diamond_full_55\u000055 tasks; strict pass requires every rubric line fully met and no penalty fires; GPT-6.1 Sol judge (medium reasoning effort); Arca chat workflow via OpenRouter; no temperature, reasoning-effort or output-token overrides, effective settings not recorded; 20 minutes per workflow turn; one answer per task\u0000Arca with GPT-6 Astra", "reported_at": "2026-11-04", "reported_date": "2026-11-04", "source_id": "lifesci_diamond_arca_blog", "source_url": "https://www.arca.inc/blog/lifesci-diamond"}, "unit": "percent"}, "series_id": "model_reports:lifesci_diamond:default", "unit": "percent"}}}, "siblings": []}