{
 "work_run": "run-2026-10",
 "work_frozen_at": "2026-10-16T00:42:50.939545+00:00",
 "std_run": "std-2026-10",
 "std_frozen_at": "2026-10-16T02:00:49.766658+00:00",
 "generated_at": "2026-10-16T02:01:31Z",
 "fx": {
  "usd_cad": 1.4263,
  "date": "2026-10-14",
  "source": "Bank of Canada daily exchange rate (FXUSDCAD)"
 },
 "models": [
  {
   "id": "rosedale-1",
   "name": "Rosedale One",
   "cls": "augure",
   "price_in_cad": 2.5,
   "price_out_cad": 6.0,
   "price_source": "Augure developer pricing"
  },
  {
   "id": "ossington-5",
   "name": "Ossington 5",
   "cls": "augure",
   "price_in_cad": 1.5,
   "price_out_cad": 3.0,
   "price_source": "Augure developer pricing"
  },
  {
   "id": "claude-sonnet-5-5",
   "name": "Claude Sonnet 5.5",
   "cls": "reference",
   "price_in_cad": 2.85,
   "price_out_cad": 14.26,
   "price_source": "https://platform.claude.com/docs/en/about-claude/pricing"
  },
  {
   "id": "gpt-5.4-mini",
   "name": "GPT-5.4 mini",
   "cls": "other",
   "price_in_cad": 1.07,
   "price_out_cad": 6.42,
   "price_source": "https://developers.openai.com/api/docs/pricing"
  },
  {
   "id": "gemini-3.8-flash",
   "name": "Gemini 3.8 Flash",
   "cls": "other",
   "price_in_cad": 1.07,
   "price_out_cad": 5.35,
   "price_source": "https://ai.google.dev/gemini-api/docs/pricing"
  },
  {
   "id": "claude-haiku-5-5",
   "name": "Claude Haiku 5.5",
   "cls": "other",
   "price_in_cad": 0.14,
   "price_out_cad": 0.71,
   "price_source": "https://platform.claude.com/docs/en/about-claude/pricing"
  },
  {
   "id": "mistral-large-4",
   "name": "Mistral Large 4",
   "cls": "other",
   "price_in_cad": 1.94,
   "price_out_cad": 5.96,
   "price_source": "https://docs.mistral.ai/models/mistral-large-4-0"
  }
 ],
 "suites": [
  {
   "key": "grounded_fr",
   "group": "work",
   "n": 60,
   "label_en": "Grounded answers · Québec sources",
   "label_fr": "Réponses ancrées · sources québécoises",
   "published": true
  },
  {
   "key": "grounded_ca",
   "group": "work",
   "n": 60,
   "label_en": "Grounded answers · Canadian law & public services",
   "label_fr": "Réponses ancrées · droit et services publics canadiens",
   "published": true
  },
  {
   "key": "ifeval_fr",
   "group": "work",
   "n": 60,
   "label_en": "French instructions · IFEval-QC",
   "label_fr": "Consignes en français · IFEval-QC",
   "published": true
  },
  {
   "key": "struct_fr",
   "group": "work",
   "n": 50,
   "label_en": "Structured extraction · Québec documents",
   "label_fr": "Extraction structurée · documents québécois",
   "published": true
  },
  {
   "key": "qc_register",
   "group": "work",
   "n": 97,
   "label_en": "Québec French register",
   "label_fr": "Registre du français québécois",
   "published": false
  },
  {
   "key": "mmlu_pro",
   "group": "general",
   "n": 500,
   "label_en": "Knowledge · MMLU-Pro",
   "label_fr": "Connaissances · MMLU-Pro",
   "published": true
  },
  {
   "key": "gpqa",
   "group": "general",
   "n": 792,
   "label_en": "Reasoning · GPQA Diamond",
   "label_fr": "Raisonnement · GPQA Diamond",
   "published": true
  },
  {
   "key": "aime2025",
   "group": "general",
   "n": 30,
   "label_en": "Math · AIME 2025",
   "label_fr": "Maths · AIME 2025",
   "published": true
  },
  {
   "key": "humaneval",
   "group": "general",
   "n": 164,
   "label_en": "Coding · HumanEval",
   "label_fr": "Code · HumanEval",
   "published": true
  }
 ],
 "rows": [
  {
   "model": "claude-haiku-5-5",
   "suite": "grounded_fr",
   "score": 98.3,
   "ci": [
    91.1,
    99.7
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 59,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "af2e20e4f0c1"
  },
  {
   "model": "claude-sonnet-5-5",
   "suite": "grounded_fr",
   "score": 98.3,
   "ci": [
    91.1,
    99.7
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 59,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "e29f115a082c"
  },
  {
   "model": "gemini-3.8-flash",
   "suite": "grounded_fr",
   "score": 98.3,
   "ci": [
    91.1,
    99.7
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 59,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "f91fc3d811c2"
  },
  {
   "model": "gpt-5.4-mini",
   "suite": "grounded_fr",
   "score": 90.0,
   "ci": [
    79.9,
    95.3
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 54,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "b1848a3df489"
  },
  {
   "model": "mistral-large-4",
   "suite": "grounded_fr",
   "score": 98.3,
   "ci": [
    91.1,
    99.7
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 59,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "74350b669870"
  },
  {
   "model": "ossington-5",
   "suite": "grounded_fr",
   "score": 98.3,
   "ci": [
    91.1,
    99.7
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 59,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "754097fabcfe"
  },
  {
   "model": "rosedale-1",
   "suite": "grounded_fr",
   "score": 100.0,
   "ci": [
    94.0,
    100.0
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 60,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "7815ee32cbd0"
  },
  {
   "model": "claude-haiku-5-5",
   "suite": "grounded_ca",
   "score": 100.0,
   "ci": [
    94.0,
    100.0
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 60,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "af5543efc005"
  },
  {
   "model": "claude-sonnet-5-5",
   "suite": "grounded_ca",
   "score": 100.0,
   "ci": [
    94.0,
    100.0
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 60,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "434eaa31d1fa"
  },
  {
   "model": "gemini-3.8-flash",
   "suite": "grounded_ca",
   "score": 100.0,
   "ci": [
    94.0,
    100.0
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 60,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "3da800c3a084"
  },
  {
   "model": "gpt-5.4-mini",
   "suite": "grounded_ca",
   "score": 98.3,
   "ci": [
    91.1,
    99.7
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 59,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "443ec462c2a8"
  },
  {
   "model": "mistral-large-4",
   "suite": "grounded_ca",
   "score": 100.0,
   "ci": [
    94.0,
    100.0
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 60,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "bd612ac7fba3"
  },
  {
   "model": "ossington-5",
   "suite": "grounded_ca",
   "score": 100.0,
   "ci": [
    94.0,
    100.0
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 60,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "71e9594f7592"
  },
  {
   "model": "rosedale-1",
   "suite": "grounded_ca",
   "score": 100.0,
   "ci": [
    94.0,
    100.0
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 60,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "e0a10fa08ca4"
  },
  {
   "model": "claude-haiku-5-5",
   "suite": "ifeval_fr",
   "score": 90.0,
   "ci": [
    79.9,
    95.3
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 54,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "57a4c7fdd04a"
  },
  {
   "model": "claude-sonnet-5-5",
   "suite": "ifeval_fr",
   "score": 95.0,
   "ci": [
    86.3,
    98.3
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 57,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "1e731d4868e9"
  },
  {
   "model": "gemini-3.8-flash",
   "suite": "ifeval_fr",
   "score": 98.3,
   "ci": [
    91.1,
    99.7
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 59,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "95623af05087"
  },
  {
   "model": "gpt-5.4-mini",
   "suite": "ifeval_fr",
   "score": 95.0,
   "ci": [
    86.3,
    98.3
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 57,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "c344572498a1"
  },
  {
   "model": "mistral-large-4",
   "suite": "ifeval_fr",
   "score": 100.0,
   "ci": [
    94.0,
    100.0
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 60,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "53a8c3e3b2d2"
  },
  {
   "model": "ossington-5",
   "suite": "ifeval_fr",
   "score": 95.0,
   "ci": [
    86.3,
    98.3
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 57,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "0a2a9a8d3f5b"
  },
  {
   "model": "rosedale-1",
   "suite": "ifeval_fr",
   "score": 90.0,
   "ci": [
    79.9,
    95.3
   ],
   "n": 60,
   "n_no_answer": 0,
   "n_correct": 54,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "03f5bda5cdff"
  },
  {
   "model": "claude-haiku-5-5",
   "suite": "struct_fr",
   "score": 98.0,
   "ci": [
    89.5,
    99.6
   ],
   "n": 50,
   "n_no_answer": 0,
   "n_correct": 49,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "23dd82dcf5dc"
  },
  {
   "model": "claude-sonnet-5-5",
   "suite": "struct_fr",
   "score": 98.0,
   "ci": [
    89.5,
    99.6
   ],
   "n": 50,
   "n_no_answer": 0,
   "n_correct": 49,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "7c5c2ffb488d"
  },
  {
   "model": "gemini-3.8-flash",
   "suite": "struct_fr",
   "score": 98.0,
   "ci": [
    89.5,
    99.6
   ],
   "n": 50,
   "n_no_answer": 0,
   "n_correct": 49,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "ce8376348719"
  },
  {
   "model": "gpt-5.4-mini",
   "suite": "struct_fr",
   "score": 98.0,
   "ci": [
    89.5,
    99.6
   ],
   "n": 50,
   "n_no_answer": 0,
   "n_correct": 49,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "68474fa86793"
  },
  {
   "model": "mistral-large-4",
   "suite": "struct_fr",
   "score": 94.0,
   "ci": [
    83.8,
    97.9
   ],
   "n": 50,
   "n_no_answer": 0,
   "n_correct": 47,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "b13c162d100d"
  },
  {
   "model": "ossington-5",
   "suite": "struct_fr",
   "score": 96.0,
   "ci": [
    86.5,
    98.9
   ],
   "n": 50,
   "n_no_answer": 0,
   "n_correct": 48,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "ee8daa55169b"
  },
  {
   "model": "rosedale-1",
   "suite": "struct_fr",
   "score": 92.0,
   "ci": [
    81.2,
    96.8
   ],
   "n": 50,
   "n_no_answer": 0,
   "n_correct": 46,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "d0da1829267c"
  },
  {
   "model": "claude-haiku-5-5",
   "suite": "qc_register",
   "score": 83.5,
   "ci": [
    74.9,
    89.6
   ],
   "n": 97,
   "n_no_answer": 0,
   "n_correct": 81,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "ddaed5c0d137"
  },
  {
   "model": "claude-sonnet-5-5",
   "suite": "qc_register",
   "score": 82.5,
   "ci": [
    73.7,
    88.8
   ],
   "n": 97,
   "n_no_answer": 0,
   "n_correct": 80,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "a09575de2d60"
  },
  {
   "model": "gemini-3.8-flash",
   "suite": "qc_register",
   "score": 85.6,
   "ci": [
    77.2,
    91.2
   ],
   "n": 97,
   "n_no_answer": 0,
   "n_correct": 83,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "40578240d24b"
  },
  {
   "model": "gpt-5.4-mini",
   "suite": "qc_register",
   "score": 91.8,
   "ci": [
    84.6,
    95.8
   ],
   "n": 97,
   "n_no_answer": 0,
   "n_correct": 89,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "06cfd63bf712"
  },
  {
   "model": "mistral-large-4",
   "suite": "qc_register",
   "score": 78.4,
   "ci": [
    69.2,
    85.4
   ],
   "n": 97,
   "n_no_answer": 0,
   "n_correct": 76,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "cd406925d3c8"
  },
  {
   "model": "ossington-5",
   "suite": "qc_register",
   "score": 67.0,
   "ci": [
    57.2,
    75.6
   ],
   "n": 97,
   "n_no_answer": 0,
   "n_correct": 65,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "d966989183cd"
  },
  {
   "model": "rosedale-1",
   "suite": "qc_register",
   "score": 71.1,
   "ci": [
    61.4,
    79.2
   ],
   "n": 97,
   "n_no_answer": 0,
   "n_correct": 69,
   "n_questions": null,
   "repeats": null,
   "ci_method": null,
   "source": "augure",
   "sha256": "b887104bcff8"
  },
  {
   "model": "claude-haiku-5-5",
   "suite": "mmlu_pro",
   "score": 87.0,
   "ci": [
    83.8,
    89.7
   ],
   "n": 500,
   "n_no_answer": 0,
   "n_correct": 435,
   "n_questions": 500,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "75dab58b2068"
  },
  {
   "model": "claude-sonnet-5-5",
   "suite": "mmlu_pro",
   "score": 89.2,
   "ci": [
    86.2,
    91.6
   ],
   "n": 500,
   "n_no_answer": 0,
   "n_correct": 446,
   "n_questions": 500,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "84681cb669ba"
  },
  {
   "model": "gemini-3.8-flash",
   "suite": "mmlu_pro",
   "score": 91.2,
   "ci": [
    88.4,
    93.4
   ],
   "n": 500,
   "n_no_answer": 0,
   "n_correct": 456,
   "n_questions": 500,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "49488c79029e"
  },
  {
   "model": "gpt-5.4-mini",
   "suite": "mmlu_pro",
   "score": 75.4,
   "ci": [
    71.4,
    79.0
   ],
   "n": 500,
   "n_no_answer": 0,
   "n_correct": 377,
   "n_questions": 500,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "e01b942a45a2"
  },
  {
   "model": "mistral-large-4",
   "suite": "mmlu_pro",
   "score": 81.6,
   "ci": [
    78.0,
    84.8
   ],
   "n": 500,
   "n_no_answer": 0,
   "n_correct": 408,
   "n_questions": 500,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "b0b678e03ec8"
  },
  {
   "model": "ossington-5",
   "suite": "mmlu_pro",
   "score": 87.8,
   "ci": [
    84.6,
    90.4
   ],
   "n": 500,
   "n_no_answer": 0,
   "n_correct": 439,
   "n_questions": 500,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "dc44deaba5e6"
  },
  {
   "model": "rosedale-1",
   "suite": "mmlu_pro",
   "score": 86.0,
   "ci": [
    82.7,
    88.8
   ],
   "n": 500,
   "n_no_answer": 2,
   "n_correct": 430,
   "n_questions": 500,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "9f4b02021b98"
  },
  {
   "model": "claude-haiku-5-5",
   "suite": "gpqa",
   "score": 84.7,
   "ci": [
    80.3,
    89.2
   ],
   "n": 792,
   "n_no_answer": 0,
   "n_correct": 671,
   "n_questions": 198,
   "repeats": 4,
   "ci_method": "clustered by question (198 questions x 4 repeats)",
   "source": "augure",
   "sha256": "9feb90cdc4fa"
  },
  {
   "model": "claude-sonnet-5-5",
   "suite": "gpqa",
   "score": 91.7,
   "ci": [
    88.3,
    95.1
   ],
   "n": 792,
   "n_no_answer": 0,
   "n_correct": 726,
   "n_questions": 198,
   "repeats": 4,
   "ci_method": "clustered by question (198 questions x 4 repeats)",
   "source": "augure",
   "sha256": "669d4fa7b076"
  },
  {
   "model": "gemini-3.8-flash",
   "suite": "gpqa",
   "score": 93.7,
   "ci": [
    90.6,
    96.8
   ],
   "n": 792,
   "n_no_answer": 0,
   "n_correct": 742,
   "n_questions": 198,
   "repeats": 4,
   "ci_method": "clustered by question (198 questions x 4 repeats)",
   "source": "augure",
   "sha256": "771ec9764e1d"
  },
  {
   "model": "gpt-5.4-mini",
   "suite": "gpqa",
   "score": 63.1,
   "ci": [
    57.5,
    68.8
   ],
   "n": 792,
   "n_no_answer": 0,
   "n_correct": 500,
   "n_questions": 198,
   "repeats": 4,
   "ci_method": "clustered by question (198 questions x 4 repeats)",
   "source": "augure",
   "sha256": "a72106e59a94"
  },
  {
   "model": "mistral-large-4",
   "suite": "gpqa",
   "score": 56.9,
   "ci": [
    50.7,
    63.2
   ],
   "n": 792,
   "n_no_answer": 0,
   "n_correct": 451,
   "n_questions": 198,
   "repeats": 4,
   "ci_method": "clustered by question (198 questions x 4 repeats)",
   "source": "augure",
   "sha256": "1a825b00265c"
  },
  {
   "model": "ossington-5",
   "suite": "gpqa",
   "score": 83.3,
   "ci": [
    79.2,
    87.4
   ],
   "n": 792,
   "n_no_answer": 33,
   "n_correct": 660,
   "n_questions": 198,
   "repeats": 4,
   "ci_method": "clustered by question (198 questions x 4 repeats)",
   "source": "augure",
   "sha256": "34689c495e4e"
  },
  {
   "model": "rosedale-1",
   "suite": "gpqa",
   "score": 83.0,
   "ci": [
    78.3,
    87.6
   ],
   "n": 792,
   "n_no_answer": 31,
   "n_correct": 657,
   "n_questions": 198,
   "repeats": 4,
   "ci_method": "clustered by question (198 questions x 4 repeats)",
   "source": "augure",
   "sha256": "0c63cd5a4731"
  },
  {
   "model": "claude-haiku-5-5",
   "suite": "aime2025",
   "score": 80.0,
   "ci": [
    62.7,
    90.5
   ],
   "n": 30,
   "n_no_answer": 0,
   "n_correct": 24,
   "n_questions": 30,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "743e2fa61bb8"
  },
  {
   "model": "claude-sonnet-5-5",
   "suite": "aime2025",
   "score": 96.7,
   "ci": [
    83.3,
    99.4
   ],
   "n": 30,
   "n_no_answer": 0,
   "n_correct": 29,
   "n_questions": 30,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "04113c3d4d4e"
  },
  {
   "model": "gemini-3.8-flash",
   "suite": "aime2025",
   "score": 93.3,
   "ci": [
    78.7,
    98.2
   ],
   "n": 30,
   "n_no_answer": 0,
   "n_correct": 28,
   "n_questions": 30,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "272d2e0189e4"
  },
  {
   "model": "gpt-5.4-mini",
   "suite": "aime2025",
   "score": 36.7,
   "ci": [
    21.9,
    54.5
   ],
   "n": 30,
   "n_no_answer": 0,
   "n_correct": 11,
   "n_questions": 30,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "441ae532de2b"
  },
  {
   "model": "mistral-large-4",
   "suite": "aime2025",
   "score": 60.0,
   "ci": [
    42.3,
    75.4
   ],
   "n": 30,
   "n_no_answer": 0,
   "n_correct": 18,
   "n_questions": 30,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "cb9fa800efc1"
  },
  {
   "model": "ossington-5",
   "suite": "aime2025",
   "score": 83.3,
   "ci": [
    66.4,
    92.7
   ],
   "n": 30,
   "n_no_answer": 2,
   "n_correct": 25,
   "n_questions": 30,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "6a1aef89cacf"
  },
  {
   "model": "rosedale-1",
   "suite": "aime2025",
   "score": 86.7,
   "ci": [
    70.3,
    94.7
   ],
   "n": 30,
   "n_no_answer": 3,
   "n_correct": 26,
   "n_questions": 30,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "cb4861a325fd"
  },
  {
   "model": "claude-haiku-5-5",
   "suite": "humaneval",
   "score": 96.3,
   "ci": [
    92.2,
    98.3
   ],
   "n": 164,
   "n_no_answer": 0,
   "n_correct": 158,
   "n_questions": 164,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "b953bc309f93"
  },
  {
   "model": "claude-sonnet-5-5",
   "suite": "humaneval",
   "score": 97.6,
   "ci": [
    93.9,
    99.0
   ],
   "n": 164,
   "n_no_answer": 0,
   "n_correct": 160,
   "n_questions": 164,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "601d995f9890"
  },
  {
   "model": "gemini-3.8-flash",
   "suite": "humaneval",
   "score": 94.5,
   "ci": [
    89.9,
    97.1
   ],
   "n": 164,
   "n_no_answer": 0,
   "n_correct": 155,
   "n_questions": 164,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "919a8bd59636"
  },
  {
   "model": "gpt-5.4-mini",
   "suite": "humaneval",
   "score": 90.2,
   "ci": [
    84.7,
    93.9
   ],
   "n": 164,
   "n_no_answer": 0,
   "n_correct": 148,
   "n_questions": 164,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "4575d656590c"
  },
  {
   "model": "mistral-large-4",
   "suite": "humaneval",
   "score": 98.2,
   "ci": [
    94.8,
    99.4
   ],
   "n": 164,
   "n_no_answer": 0,
   "n_correct": 161,
   "n_questions": 164,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "8f7111649275"
  },
  {
   "model": "ossington-5",
   "suite": "humaneval",
   "score": 97.6,
   "ci": [
    93.9,
    99.0
   ],
   "n": 164,
   "n_no_answer": 0,
   "n_correct": 160,
   "n_questions": 164,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "f06c58d579ef"
  },
  {
   "model": "rosedale-1",
   "suite": "humaneval",
   "score": 98.2,
   "ci": [
    94.8,
    99.4
   ],
   "n": 164,
   "n_no_answer": 0,
   "n_correct": 161,
   "n_questions": 164,
   "repeats": 1,
   "ci_method": "wilson",
   "source": "augure",
   "sha256": "a9ccdedf5c68"
  }
 ]
}