[
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "2025_human_panel",
    "score": 0.98,
    "costPerTask": 17,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "2025_human_panel",
    "score": 1,
    "costPerTask": 17,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "2025_human_panel",
    "score": 1,
    "costPerTask": 17,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Private_Eval",
    "modelId": "2025_human_panel",
    "score": 1,
    "costPerTask": 17,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "ARChitects",
    "score": 0.56,
    "costPerTask": 0.2,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "ARChitects",
    "score": 0.035,
    "costPerTask": 0.2,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "NVARC_2025",
    "score": 0.2764,
    "costPerTask": 0.2,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "ARChitects",
    "score": 0.025,
    "costPerTask": 0.2,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Private_Eval",
    "modelId": "ARChitects",
    "score": 0.031,
    "costPerTask": 0.2,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "Claude 3.7",
    "score": 0.136,
    "costPerTask": 0.058,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "Claude 3.7",
    "score": 0,
    "costPerTask": 0.12,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "Claude 3.7",
    "score": 0,
    "costPerTask": 0.12,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "Claude 3.7 Thinking 16K",
    "score": 0.286,
    "costPerTask": 0.33,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "Claude 3.7 Thinking 16K",
    "score": 0.008,
    "costPerTask": 0.57,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "Claude 3.7 Thinking 16K",
    "score": 0.007,
    "costPerTask": 0.51,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "Claude 3.7 Thinking 1K",
    "score": 0.116,
    "costPerTask": 0.07,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "Claude 3.7 Thinking 1K",
    "score": 0.008,
    "costPerTask": 0.14,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "Claude 3.7 Thinking 1K",
    "score": 0.004,
    "costPerTask": 0.14,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "Claude 3.7 Thinking 8K",
    "score": 0.212,
    "costPerTask": 0.21,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "Claude 3.7 Thinking 8K",
    "score": 0,
    "costPerTask": 0.36,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "Claude 3.7 Thinking 8K",
    "score": 0.009,
    "costPerTask": 0.36,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "Gemini 1.5 Pro",
    "score": 0,
    "costPerTask": 0.04,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "Gemini 1.5 Pro",
    "score": 0.008,
    "costPerTask": 0.04,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "Gemini 2.0 Flash",
    "score": 0.013,
    "costPerTask": 0.004,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-4-5-2025-02-27",
    "score": 0.103,
    "costPerTask": 0.29,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-4-5-2025-02-27",
    "score": 0,
    "costPerTask": 2.07,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-4-5-2025-02-27",
    "score": 0.008,
    "costPerTask": 2.1,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-4o-2024-11-20",
    "score": 0.045,
    "costPerTask": 0.05,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-4o-2024-11-20",
    "score": 0,
    "costPerTask": 0.08,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-4o-2024-11-20",
    "score": 0,
    "costPerTask": 0.08,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-4o-mini-2024-07-18",
    "score": 0,
    "costPerTask": 0.01,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-4o-mini-2024-07-18",
    "score": 0,
    "costPerTask": 0.01,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "Icecuber",
    "score": 0.17,
    "costPerTask": 0.2,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "Icecuber",
    "score": 0.029,
    "costPerTask": 0.13,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "Icecuber",
    "score": 0.016,
    "costPerTask": 0.13,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Private_Eval",
    "modelId": "Icecuber",
    "score": 0.004,
    "costPerTask": 0.13,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "mturker",
    "score": 0.77,
    "costPerTask": 3,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o3-preview-low",
    "score": 0.757,
    "costPerTask": 200,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o3-preview-low",
    "score": 0.04,
    "costPerTask": 200,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "R1",
    "score": 0.158,
    "costPerTask": 0.06,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "R1",
    "score": 0.003,
    "costPerTask": 0.08,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "R1",
    "score": 0.013,
    "costPerTask": 0.08,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "stem_grad",
    "score": 0.98,
    "costPerTask": 10,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "Llama-4-Maverick-17B-128E-Instruct-FP8-together",
    "score": 0.0438,
    "costPerTask": 0.0078,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "Llama-4-Maverick-17B-128E-Instruct-FP8-together",
    "score": 0.0712,
    "costPerTask": 0.0065,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "Llama-4-Maverick-17B-128E-Instruct-FP8-together",
    "score": 0,
    "costPerTask": 0.0126,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "Llama-4-Maverick-17B-128E-Instruct-FP8-together",
    "score": 0,
    "costPerTask": 0.0121,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "Llama-4-Scout-17B-16E-Instruct-together",
    "score": 0.005,
    "costPerTask": 0.0041,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "Llama-4-Scout-17B-16E-Instruct-together",
    "score": 0.0238,
    "costPerTask": 0.0036,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "Llama-4-Scout-17B-16E-Instruct-together",
    "score": 0,
    "costPerTask": 0.006,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "Llama-4-Scout-17B-16E-Instruct-together",
    "score": 0,
    "costPerTask": 0.0062,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-4-1-nano-2025-04-14",
    "score": 0,
    "costPerTask": 0.0021,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-4-1-nano-2025-04-14",
    "score": 0.0175,
    "costPerTask": 0.0018,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-4-1-nano-2025-04-14",
    "score": 0,
    "costPerTask": 0.0036,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-4-1-nano-2025-04-14",
    "score": 0,
    "costPerTask": 0.0036,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-4-1-mini-2025-04-14",
    "score": 0.035,
    "costPerTask": 0.0078,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-4-1-mini-2025-04-14",
    "score": 0.0725,
    "costPerTask": 0.0066,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-4-1-mini-2025-04-14",
    "score": 0,
    "costPerTask": 0.0136,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-4-1-mini-2025-04-14",
    "score": 0,
    "costPerTask": 0.0139,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-4-1-2025-04-14",
    "score": 0.055,
    "costPerTask": 0.039,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-4-1-2025-04-14",
    "score": 0.1175,
    "costPerTask": 0.032,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-4-1-2025-04-14",
    "score": 0,
    "costPerTask": 0.069,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-4-1-2025-04-14",
    "score": 0.0042,
    "costPerTask": 0.0691,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o1-mini-2024-09-12",
    "score": 0.1316,
    "costPerTask": 0.1239,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o1-mini-2024-09-12",
    "score": 0.14,
    "costPerTask": 0.135,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o1-mini-2024-09-12",
    "score": 0.0,
    "costPerTask": 0.1838,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o1-mini-2024-09-12",
    "score": 0.0083,
    "costPerTask": 0.1907,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o3-mini-2025-01-31-medium",
    "score": 0.3063,
    "costPerTask": 0.165,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o3-mini-2025-01-31-low",
    "score": 0.1738,
    "costPerTask": 0.0442,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o3-mini-2025-01-31-low",
    "score": 0.145,
    "costPerTask": 0.0519,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o3-mini-2025-01-31-low",
    "score": 0.0,
    "costPerTask": 0.0596,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o3-mini-2025-01-31-low",
    "score": 0.0,
    "costPerTask": 0.0623,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o3-mini-2025-01-31-medium",
    "score": 0.2233,
    "costPerTask": 0.1907,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o3-mini-2025-01-31-medium",
    "score": 0.0,
    "costPerTask": 0.2723,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o3-mini-2025-01-31-medium",
    "score": 0.0208,
    "costPerTask": 0.2843,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o3-mini-2025-01-31-high",
    "score": 0.4658,
    "costPerTask": 0.3048,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o3-mini-2025-01-31-high",
    "score": 0.345,
    "costPerTask": 0.3989,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o3-mini-2025-01-31-high",
    "score": 0.0,
    "costPerTask": 0.6072,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o3-mini-2025-01-31-high",
    "score": 0.0299,
    "costPerTask": 0.5472,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o3-2025-04-16-low",
    "score": 0.4761,
    "costPerTask": 0.1591,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o3-2025-04-16-low",
    "score": 0.415,
    "costPerTask": 0.1764,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o3-2025-04-16-low",
    "score": 0.0266,
    "costPerTask": 0.245,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o3-2025-04-16-low",
    "score": 0.0199,
    "costPerTask": 0.2343,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o3-2025-04-16-medium",
    "score": 0.5671,
    "costPerTask": 0.2613,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o3-2025-04-16-medium",
    "score": 0.5383,
    "costPerTask": 0.2882,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o3-2025-04-16-medium",
    "score": 0.0449,
    "costPerTask": 0.4994,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o3-2025-04-16-medium",
    "score": 0.0298,
    "costPerTask": 0.4787,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o3-2025-04-16-high",
    "score": 0.6425,
    "costPerTask": 0.4041,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o3-2025-04-16-high",
    "score": 0.6083,
    "costPerTask": 0.5002,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o3-2025-04-16-high",
    "score": 0.0292,
    "costPerTask": 0.9001,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o3-2025-04-16-high",
    "score": 0.0653,
    "costPerTask": 0.8339,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o4-mini-2025-04-16-low",
    "score": 0.2762,
    "costPerTask": 0.0359,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o4-mini-2025-04-16-low",
    "score": 0.2133,
    "costPerTask": 0.0406,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o4-mini-2025-04-16-low",
    "score": 0.0028,
    "costPerTask": 0.0518,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o4-mini-2025-04-16-low",
    "score": 0.0167,
    "costPerTask": 0.05,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o4-mini-2025-04-16-medium",
    "score": 0.5025,
    "costPerTask": 0.129,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o4-mini-2025-04-16-medium",
    "score": 0.4183,
    "costPerTask": 0.15,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o4-mini-2025-04-16-medium",
    "score": 0.0222,
    "costPerTask": 0.237,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o4-mini-2025-04-16-medium",
    "score": 0.0236,
    "costPerTask": 0.2311,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o4-mini-2025-04-16-high",
    "score": 0.6803,
    "costPerTask": 0.3246,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o4-mini-2025-04-16-high",
    "score": 0.5867,
    "costPerTask": 0.4058,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o4-mini-2025-04-16-high",
    "score": 0.0752,
    "costPerTask": 0.8822,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o4-mini-2025-04-16-high",
    "score": 0.0611,
    "costPerTask": 0.856,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-flash-preview-05-20",
    "score": 0.363,
    "costPerTask": 0.033,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-flash-preview-05-20",
    "score": 0.3333,
    "costPerTask": 0.0371,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-flash-preview-05-20",
    "score": 0.0194,
    "costPerTask": 0.0586,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-flash-preview-05-20",
    "score": 0.0169,
    "costPerTask": 0.057,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-1k",
    "score": 0.1796,
    "costPerTask": 0.0231,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-1k",
    "score": 0.16,
    "costPerTask": 0.0356,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-1k",
    "score": 0.0,
    "costPerTask": 0.0348,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-1k",
    "score": 0.0216,
    "costPerTask": 0.0302,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-8k",
    "score": 0.3453,
    "costPerTask": 0.1158,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-8k",
    "score": 0.2583,
    "costPerTask": 0.1344,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-8k",
    "score": 0.0111,
    "costPerTask": 0.2047,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-8k",
    "score": 0.0212,
    "costPerTask": 0.1994,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-16k",
    "score": 0.36,
    "costPerTask": 0.1846,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-16k",
    "score": 0.3333,
    "costPerTask": 0.2134,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-16k",
    "score": 0.0111,
    "costPerTask": 0.324,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-16k",
    "score": 0.0198,
    "costPerTask": 0.3173,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-24k",
    "score": 0.3912,
    "costPerTask": 0.1858,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-24k",
    "score": 0.3233,
    "costPerTask": 0.1971,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-24k",
    "score": 0.0264,
    "costPerTask": 0.3305,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-flash-preview-05-20-thinking-24k",
    "score": 0.0254,
    "costPerTask": 0.3191,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "codex-mini-latest",
    "score": 0.3338,
    "costPerTask": 0.1269,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "codex-mini-latest",
    "score": 0.2733,
    "costPerTask": 0.1597,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "codex-mini-latest",
    "score": 0.0,
    "costPerTask": 0.2272,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "codex-mini-latest",
    "score": 0.0127,
    "costPerTask": 0.23,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-sonnet-4-20250514",
    "score": 0.33,
    "costPerTask": 0.0696,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-sonnet-4-20250514",
    "score": 0.2383,
    "costPerTask": 0.0806,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-sonnet-4-20250514",
    "score": 0.0208,
    "costPerTask": 0.1311,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-sonnet-4-20250514",
    "score": 0.0127,
    "costPerTask": 0.1272,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-sonnet-4-20250514-thinking-1k",
    "score": 0.313,
    "costPerTask": 0.0811,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-sonnet-4-20250514-thinking-1k",
    "score": 0.28,
    "costPerTask": 0.0937,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-sonnet-4-20250514-thinking-1k",
    "score": 0.0125,
    "costPerTask": 0.1466,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-sonnet-4-20250514-thinking-1k",
    "score": 0.0085,
    "costPerTask": 0.1425,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-sonnet-4-20250514-thinking-8k-bedrock",
    "score": 0.4862,
    "costPerTask": 0.1936,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-sonnet-4-20250514-thinking-8k-bedrock",
    "score": 0.29,
    "costPerTask": 0.1952,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-sonnet-4-20250514-thinking-8k-bedrock",
    "score": 0.0292,
    "costPerTask": 0.2724,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-sonnet-4-20250514-thinking-8k-bedrock",
    "score": 0.0212,
    "costPerTask": 0.2654,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-sonnet-4-20250514-thinking-16k-bedrock",
    "score": 0.5675,
    "costPerTask": 0.322,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-sonnet-4-20250514-thinking-16k-bedrock",
    "score": 0.4,
    "costPerTask": 0.3658,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-sonnet-4-20250514-thinking-16k-bedrock",
    "score": 0.0444,
    "costPerTask": 0.5102,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-sonnet-4-20250514-thinking-16k-bedrock",
    "score": 0.0593,
    "costPerTask": 0.4857,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-20250514-thinking-16k",
    "score": 0.5425,
    "costPerTask": 1.1405,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-20250514-thinking-16k",
    "score": 0.3567,
    "costPerTask": 1.2496,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-20250514-thinking-16k",
    "score": 0.0333,
    "costPerTask": 2.0182,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-20250514-thinking-16k",
    "score": 0.0861,
    "costPerTask": 1.9284,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-20250514-thinking-8k",
    "score": 0.4562,
    "costPerTask": 0.7817,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-20250514-thinking-8k",
    "score": 0.3067,
    "costPerTask": 0.7408,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-20250514-thinking-8k",
    "score": 0.025,
    "costPerTask": 1.2021,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-20250514-thinking-8k",
    "score": 0.0452,
    "costPerTask": 1.1569,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-20250514-thinking-1k",
    "score": 0.4331,
    "costPerTask": 0.4236,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-20250514-thinking-1k",
    "score": 0.27,
    "costPerTask": 0.5021,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-20250514-thinking-1k",
    "score": 0.0043,
    "costPerTask": 0.7578,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-20250514-thinking-1k",
    "score": 0.0,
    "costPerTask": 0.7503,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-20250514",
    "score": 0.355,
    "costPerTask": 0.3485,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-20250514",
    "score": 0.225,
    "costPerTask": 0.4036,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-20250514",
    "score": 0.0125,
    "costPerTask": 0.6634,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-20250514",
    "score": 0.0127,
    "costPerTask": 0.6388,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "deepseek_r1_0528-openrouter",
    "score": 0.2698,
    "costPerTask": 0.047,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "deepseek_r1_0528-openrouter",
    "score": 0.2121,
    "costPerTask": 0.0464,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "deepseek_r1_0528-openrouter",
    "score": 0.0029,
    "costPerTask": 0.0495,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "deepseek_r1_0528-openrouter",
    "score": 0.0112,
    "costPerTask": 0.0527,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "grok-3-openrouter",
    "score": 0.0838,
    "costPerTask": 0.0735,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "grok-3-openrouter",
    "score": 0.055,
    "costPerTask": 0.0931,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "grok-3-openrouter",
    "score": 0.0,
    "costPerTask": 0.1402,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "grok-3-openrouter",
    "score": 0.0,
    "costPerTask": 0.1421,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "grok-3-mini-beta-low-openrouter",
    "score": 0.1762,
    "costPerTask": 0.0087,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "grok-3-mini-beta-low-openrouter",
    "score": 0.165,
    "costPerTask": 0.0099,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "grok-3-mini-beta-low-openrouter",
    "score": 0.0,
    "costPerTask": 0.013,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "grok-3-mini-beta-low-openrouter",
    "score": 0.0042,
    "costPerTask": 0.0131,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-pro-preview-openrouter",
    "score": 0.43,
    "costPerTask": 0.5032,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-pro-preview-openrouter",
    "score": 0.33,
    "costPerTask": 0.5694,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-pro-preview-openrouter",
    "score": 0.0111,
    "costPerTask": 0.8233,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-pro-preview-openrouter",
    "score": 0.0375,
    "costPerTask": 0.8126,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-pro-preview-05-06-thinking-1k",
    "score": 0.4819,
    "costPerTask": 0.4802,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-pro-preview-05-06-thinking-1k",
    "score": 0.3133,
    "costPerTask": 0.5866,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-pro-preview-05-06-thinking-1k",
    "score": 0.0133,
    "costPerTask": 0.7947,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-pro-preview-05-06-thinking-1k",
    "score": 0.0344,
    "costPerTask": 0.8037,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o3-pro-2025-06-10-low",
    "score": 0.5088,
    "costPerTask": 1.5093,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o3-pro-2025-06-10-low",
    "score": 0.4433,
    "costPerTask": 1.6382,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o3-pro-2025-06-10-low",
    "score": 0.0194,
    "costPerTask": 2.455,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o3-pro-2025-06-10-low",
    "score": 0.0205,
    "costPerTask": 2.2293,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o3-pro-2025-06-10-medium",
    "score": 0.5806,
    "costPerTask": 2.546,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o3-pro-2025-06-10-medium",
    "score": 0.57,
    "costPerTask": 3.1766,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o3-pro-2025-06-10-medium",
    "score": 0.0347,
    "costPerTask": 5.1645,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o3-pro-2025-06-10-medium",
    "score": 0.0194,
    "costPerTask": 4.7441,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "o3-pro-2025-06-10-high",
    "score": 0.6334,
    "costPerTask": 3.9228,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "o3-pro-2025-06-10-high",
    "score": 0.5933,
    "costPerTask": 4.16,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "o3-pro-2025-06-10-high",
    "score": 0.0389,
    "costPerTask": 9.1492,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "o3-pro-2025-06-10-high",
    "score": 0.0486,
    "costPerTask": 7.5516,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "magistral-small-2506",
    "score": 0.0862,
    "costPerTask": 0.0289,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "magistral-small-2506",
    "score": 0.05,
    "costPerTask": 0.0399,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "magistral-small-2506",
    "score": 0,
    "costPerTask": 0.0495,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "magistral-small-2506",
    "score": 0,
    "costPerTask": 0.0488,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "magistral-medium-2506",
    "score": 0.089,
    "costPerTask": 0.1069,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "magistral-medium-2506",
    "score": 0.0591,
    "costPerTask": 0.1015,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "magistral-medium-2506",
    "score": 0,
    "costPerTask": 0.1061,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "magistral-medium-2506",
    "score": 0,
    "costPerTask": 0.1079,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "magistral-medium-2506-thinking",
    "score": 0.0799,
    "costPerTask": 0.1059,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "magistral-medium-2506-thinking",
    "score": 0.0612,
    "costPerTask": 0.0989,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "magistral-medium-2506-thinking",
    "score": 0,
    "costPerTask": 0.1242,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "magistral-medium-2506-thinking",
    "score": 0,
    "costPerTask": 0.123,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-1k",
    "score": 0.175,
    "costPerTask": 0.048,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-1k",
    "score": 0.16,
    "costPerTask": 0.0573,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-1k",
    "score": 0.0042,
    "costPerTask": 0.0879,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-1k",
    "score": 0,
    "costPerTask": 0.0885,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-8k",
    "score": 0.4419,
    "costPerTask": 0.2605,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-8k",
    "score": 0.295,
    "costPerTask": 0.2947,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-8k",
    "score": 0.0264,
    "costPerTask": 0.4493,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-8k",
    "score": 0.0292,
    "costPerTask": 0.4439,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-16k",
    "score": 0.5643,
    "costPerTask": 0.3964,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-16k",
    "score": 0.41,
    "costPerTask": 0.4839,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-16k",
    "score": 0.0514,
    "costPerTask": 0.7222,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-16k",
    "score": 0.0403,
    "costPerTask": 0.7145,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-32k",
    "score": 0.5588,
    "costPerTask": 0.4164,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-32k",
    "score": 0.37,
    "costPerTask": 0.5123,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-32k",
    "score": 0.0458,
    "costPerTask": 0.7628,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-2-5-pro-2025-06-17-thinking-32k",
    "score": 0.0486,
    "costPerTask": 0.757,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "grok-4-0709",
    "score": 0.7374,
    "costPerTask": 0.8179,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "grok-4-0709",
    "score": 0.6667,
    "costPerTask": 1.0136,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "grok-4-0709",
    "score": 0.2115,
    "costPerTask": 2.0759,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "grok-4-0709",
    "score": 0.15975,
    "costPerTask": 2.1659,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "qwen3-235b-a22b-instruct-2507",
    "score": 0.17,
    "costPerTask": 0.0021,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "qwen3-235b-a22b-instruct-2507",
    "score": 0.11,
    "costPerTask": 0.0025,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "qwen3-235b-a22b-instruct-2507",
    "score": 0.0083,
    "costPerTask": 0.0045,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "qwen3-235b-a22b-instruct-2507",
    "score": 0.0125,
    "costPerTask": 0.0044,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2025-08-07-high",
    "score": 0.6588,
    "costPerTask": 0.3924,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2025-08-07-high",
    "score": 0.6567,
    "costPerTask": 0.5087,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2025-08-07-high",
    "score": 0.0958,
    "costPerTask": 0.774,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2025-08-07-high",
    "score": 0.0986,
    "costPerTask": 0.7302,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2025-08-07-medium",
    "score": 0.634,
    "costPerTask": 0.2478,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2025-08-07-medium",
    "score": 0.5617,
    "costPerTask": 0.3301,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2025-08-07-medium",
    "score": 0.0763,
    "costPerTask": 0.4753,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2025-08-07-medium",
    "score": 0.0749,
    "costPerTask": 0.4486,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2025-08-07-low",
    "score": 0.4837,
    "costPerTask": 0.1296,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2025-08-07-low",
    "score": 0.44,
    "costPerTask": 0.1531,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2025-08-07-low",
    "score": 0.025,
    "costPerTask": 0.2001,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2025-08-07-low",
    "score": 0.0194,
    "costPerTask": 0.1896,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2025-08-07-low",
    "score": 0.0194,
    "costPerTask": 0.1896,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2025-08-07-minimal",
    "score": 0.1115,
    "costPerTask": 0.0244,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2025-08-07-minimal",
    "score": 0.06,
    "costPerTask": 0.0335,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2025-08-07-minimal",
    "score": 0,
    "costPerTask": 0.0543,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2025-08-07-minimal",
    "score": 0,
    "costPerTask": 0.0562,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-mini-2025-08-07-high",
    "score": 0.6152,
    "costPerTask": 0.0936,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-mini-2025-08-07-high",
    "score": 0.5433,
    "costPerTask": 0.116,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-mini-2025-08-07-high",
    "score": 0.0583,
    "costPerTask": 0.215,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-mini-2025-08-07-high",
    "score": 0.0444,
    "costPerTask": 0.1977,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-mini-2025-08-07-medium",
    "score": 0.4625,
    "costPerTask": 0.039,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-mini-2025-08-07-medium",
    "score": 0.3733,
    "costPerTask": 0.0401,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-mini-2025-08-07-medium",
    "score": 0.0056,
    "costPerTask": 0.0674,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-mini-2025-08-07-medium",
    "score": 0.0403,
    "costPerTask": 0.0629,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-mini-2025-08-07-low",
    "score": 0.2438,
    "costPerTask": 0.0136,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-mini-2025-08-07-low",
    "score": 0.2633,
    "costPerTask": 0.0135,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-mini-2025-08-07-low",
    "score": 0.0028,
    "costPerTask": 0.0193,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-mini-2025-08-07-low",
    "score": 0.0083,
    "costPerTask": 0.0189,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-mini-2025-08-07-minimal",
    "score": 0.0714,
    "costPerTask": 0.0051,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-mini-2025-08-07-minimal",
    "score": 0.0533,
    "costPerTask": 0.0057,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-mini-2025-08-07-minimal",
    "score": 0,
    "costPerTask": 0.0087,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-mini-2025-08-07-minimal",
    "score": 0.0167,
    "costPerTask": 0.0094,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-nano-2025-08-07-high",
    "score": 0.2967,
    "costPerTask": 0.0245,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-nano-2025-08-07-high",
    "score": 0.1667,
    "costPerTask": 0.0292,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-nano-2025-08-07-high",
    "score": 0.003,
    "costPerTask": 0.0321,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-nano-2025-08-07-high",
    "score": 0.0261,
    "costPerTask": 0.0295,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-nano-2025-08-07-medium",
    "score": 0.2082,
    "costPerTask": 0.0119,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-nano-2025-08-07-medium",
    "score": 0.2071,
    "costPerTask": 0.0124,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-nano-2025-08-07-medium",
    "score": 0,
    "costPerTask": 0.0147,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-nano-2025-08-07-medium",
    "score": 0.0088,
    "costPerTask": 0.0137,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-nano-2025-08-07-low",
    "score": 0.1181,
    "costPerTask": 0.0031,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-nano-2025-08-07-low",
    "score": 0.0404,
    "costPerTask": 0.0033,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-nano-2025-08-07-low",
    "score": 0,
    "costPerTask": 0.0035,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-nano-2025-08-07-low",
    "score": 0,
    "costPerTask": 0.0033,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-nano-2025-08-07-minimal",
    "score": 0.0275,
    "costPerTask": 0.0012,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-nano-2025-08-07-minimal",
    "score": 0.015,
    "costPerTask": 0.0015,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-nano-2025-08-07-minimal",
    "score": 0,
    "costPerTask": 0.0026,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-nano-2025-08-07-minimal",
    "score": 0,
    "costPerTask": 0.0025,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "hrm-2-2508",
    "score": 0.32,
    "costPerTask": 1.48,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "hrm-2-2508",
    "score": 0.02,
    "costPerTask": 1.68,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "jeremy_sept_2025",
    "score": 0.796,
    "costPerTask": 8.42,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "jeremy_sept_2025",
    "score": 0.2944,
    "costPerTask": 30.4,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "eric_pang_sept_2025",
    "score": 0.771,
    "costPerTask": 2.56,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "eric_pang_sept_2025",
    "score": 0.26,
    "costPerTask": 3.97,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-sonnet-4-5-20250929",
    "score": 0.3538,
    "costPerTask": 0.0691,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-sonnet-4-5-20250929",
    "score": 0.255,
    "costPerTask": 0.0808,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-sonnet-4-5-20250929",
    "score": 0.0375,
    "costPerTask": 0.1313,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-sonnet-4-5-20250929",
    "score": 0.0375,
    "costPerTask": 0.1295,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-sonnet-4-5-20250929-thinking-1k",
    "score": 0.3662,
    "costPerTask": 0.0823,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-sonnet-4-5-20250929-thinking-1k",
    "score": 0.31,
    "costPerTask": 0.0928,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-sonnet-4-5-20250929-thinking-1k",
    "score": 0.0208,
    "costPerTask": 0.1424,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-sonnet-4-5-20250929-thinking-1k",
    "score": 0.0583,
    "costPerTask": 0.1424,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-sonnet-4-5-20250929-thinking-8k",
    "score": 0.535,
    "costPerTask": 0.1681,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-sonnet-4-5-20250929-thinking-8k",
    "score": 0.465,
    "costPerTask": 0.1776,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-sonnet-4-5-20250929-thinking-8k",
    "score": 0.0486,
    "costPerTask": 0.2409,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-sonnet-4-5-20250929-thinking-8k",
    "score": 0.0694,
    "costPerTask": 0.2349,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-sonnet-4-5-20250929-thinking-16k",
    "score": 0.6362,
    "costPerTask": 0.2423,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-sonnet-4-5-20250929-thinking-16k",
    "score": 0.4833,
    "costPerTask": 0.2683,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-sonnet-4-5-20250929-thinking-16k",
    "score": 0.0653,
    "costPerTask": 0.3557,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-sonnet-4-5-20250929-thinking-16k",
    "score": 0.0694,
    "costPerTask": 0.3499,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-sonnet-4-5-20250929-thinking-32k",
    "score": 0.7375,
    "costPerTask": 0.4419,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-sonnet-4-5-20250929-thinking-32k",
    "score": 0.6367,
    "costPerTask": 0.5193,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-sonnet-4-5-20250929-thinking-32k",
    "score": 0.1472,
    "costPerTask": 0.8403,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-sonnet-4-5-20250929-thinking-32k",
    "score": 0.1361,
    "costPerTask": 0.7589,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-pro-2025-10-06",
    "score": 0.77,
    "costPerTask": 4.0316,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-pro-2025-10-06",
    "score": 0.7017,
    "costPerTask": 4.7785,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-pro-2025-10-06",
    "score": 0.1333,
    "costPerTask": 8.0107,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-pro-2025-10-06",
    "score": 0.1833,
    "costPerTask": 7.1432,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "trm-2025-10-07",
    "score": 0.44,
    "costPerTask": 0.39465,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "trm-2025-10-07",
    "score": 0.4,
    "costPerTask": 1.7638,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "trm-2025-10-07",
    "score": 0.05,
    "costPerTask": 1.8,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "trm-2025-10-07",
    "score": 0.0625,
    "costPerTask": 2.1,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-haiku-4-5-20251001",
    "score": 0.2662,
    "costPerTask": 0.0223,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-haiku-4-5-20251001",
    "score": 0.1433,
    "costPerTask": 0.0263,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-haiku-4-5-20251001",
    "score": 0,
    "costPerTask": 0.0432,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-haiku-4-5-20251001",
    "score": 0.0125,
    "costPerTask": 0.0426,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-haiku-4-5-20251001-thinking-1k",
    "score": 0.2712,
    "costPerTask": 0.0268,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-haiku-4-5-20251001-thinking-1k",
    "score": 0.1683,
    "costPerTask": 0.0306,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-haiku-4-5-20251001-thinking-1k",
    "score": 0,
    "costPerTask": 0.0478,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-haiku-4-5-20251001-thinking-1k",
    "score": 0.0125,
    "costPerTask": 0.0471,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-haiku-4-5-20251001-thinking-8k",
    "score": 0.45,
    "costPerTask": 0.0644,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-haiku-4-5-20251001-thinking-8k",
    "score": 0.255,
    "costPerTask": 0.0687,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-haiku-4-5-20251001-thinking-8k",
    "score": 0.0208,
    "costPerTask": 0.0952,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-haiku-4-5-20251001-thinking-8k",
    "score": 0.0167,
    "costPerTask": 0.0909,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-haiku-4-5-20251001-thinking-16k",
    "score": 0.5138,
    "costPerTask": 0.0979,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-haiku-4-5-20251001-thinking-16k",
    "score": 0.3733,
    "costPerTask": 0.1042,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-haiku-4-5-20251001-thinking-16k",
    "score": 0.0417,
    "costPerTask": 0.1432,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-haiku-4-5-20251001-thinking-16k",
    "score": 0.0278,
    "costPerTask": 0.1392,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-haiku-4-5-20251001-thinking-32k",
    "score": 0.6288,
    "costPerTask": 0.2139,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-haiku-4-5-20251001-thinking-32k",
    "score": 0.4767,
    "costPerTask": 0.2558,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-haiku-4-5-20251001-thinking-32k",
    "score": 0.0514,
    "costPerTask": 0.3927,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-haiku-4-5-20251001-thinking-32k",
    "score": 0.0403,
    "costPerTask": 0.3766,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "grok-4-fast-reasoning",
    "score": 0.5813,
    "costPerTask": 0.0278,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "grok-4-fast-reasoning",
    "score": 0.485,
    "costPerTask": 0.0307,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "grok-4-fast-reasoning",
    "score": 0.0458,
    "costPerTask": 0.0614,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "grok-4-fast-reasoning",
    "score": 0.0528,
    "costPerTask": 0.061,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-1-2025-11-13-thinking-none",
    "score": 0.1238,
    "costPerTask": 0.0275,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-1-2025-11-13-thinking-none",
    "score": 0,
    "costPerTask": 0.0591,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-1-2025-11-13-thinking-none",
    "score": 0.0583,
    "costPerTask": 0.0338,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-1-2025-11-13-thinking-none",
    "score": 0.0042,
    "costPerTask": 0.0581,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-1-2025-11-13-thinking-low",
    "score": 0.44,
    "costPerTask": 0.0929,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-1-2025-11-13-thinking-low",
    "score": 0.0222,
    "costPerTask": 0.1348,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-1-2025-11-13-thinking-low",
    "score": 0.3317,
    "costPerTask": 0.1,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-1-2025-11-13-thinking-low",
    "score": 0.0194,
    "costPerTask": 0.129,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-1-2025-11-13-thinking-medium",
    "score": 0.6886,
    "costPerTask": 0.2563,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-1-2025-11-13-thinking-medium",
    "score": 0.084,
    "costPerTask": 0.4599,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-1-2025-11-13-thinking-medium",
    "score": 0.5767,
    "costPerTask": 0.3149,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-1-2025-11-13-thinking-medium",
    "score": 0.0653,
    "costPerTask": 0.4206,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-1-2025-11-13-thinking-high",
    "score": 0.7712,
    "costPerTask": 0.546,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-1-2025-11-13-thinking-high",
    "score": 0.1833,
    "costPerTask": 1.2166,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-1-2025-11-13-thinking-high",
    "score": 0.7283,
    "costPerTask": 0.674,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-1-2025-11-13-thinking-high",
    "score": 0.1764,
    "costPerTask": 1.1704,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-3-pro-preview",
    "score": 0.75,
    "costPerTask": 0.4926,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-3-pro-preview",
    "score": 0.3111,
    "costPerTask": 0.8109,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-3-deep-think-preview",
    "score": 0.875,
    "costPerTask": 44.25900135,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-3-deep-think-preview",
    "score": 0.4514,
    "costPerTask": 77.16309638,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-5-20251101-thinking-none",
    "score": 0.5261,
    "costPerTask": 0.11,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-5-20251101-thinking-none",
    "score": 0.0708,
    "costPerTask": 0.22,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-none",
    "score": 0.4,
    "costPerTask": 0.13,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-none",
    "score": 0.0778,
    "costPerTask": 0.22,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-5-20251101-thinking-1k",
    "score": 0.5132,
    "costPerTask": 0.13,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-5-20251101-thinking-1k",
    "score": 0.05,
    "costPerTask": 0.23,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-1k",
    "score": 0.3517,
    "costPerTask": 0.15,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-1k",
    "score": 0.0944,
    "costPerTask": 0.23,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-5-20251101-thinking-8k",
    "score": 0.701,
    "costPerTask": 0.34,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-5-20251101-thinking-8k",
    "score": 0.1042,
    "costPerTask": 0.49,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-8k",
    "score": 0.5867,
    "costPerTask": 0.38,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-8k",
    "score": 0.1389,
    "costPerTask": 0.48,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-5-20251101-thinking-16k",
    "score": 0.8161,
    "costPerTask": 0.52,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-5-20251101-thinking-16k",
    "score": 0.2417,
    "costPerTask": 0.82,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-16k",
    "score": 0.72,
    "costPerTask": 0.62,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-16k",
    "score": 0.2278,
    "costPerTask": 0.79,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-5-20251101-thinking-32k",
    "score": 0.8662,
    "costPerTask": 0.76,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-5-20251101-thinking-32k",
    "score": 0.2815,
    "costPerTask": 1.37,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-32k",
    "score": 0.7583,
    "costPerTask": 0.95,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-32k",
    "score": 0.3056,
    "costPerTask": 1.29,
    "resultsUrl": "",
    "display": false
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-64k",
    "score": 0.8,
    "costPerTask": 1.47,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-5-20251101-thinking-64k",
    "score": 0.3764,
    "costPerTask": 2.4,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "poetiq-arc-agi-gemini-3-d_v1",
    "score": 0.54,
    "costPerTask": 30.568,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2-2025-12-11-thinking-none",
    "score": 0.165,
    "costPerTask": 0.0385,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2-2025-12-11-thinking-none",
    "score": 0,
    "costPerTask": 0.0824,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2-2025-12-11-thinking-none",
    "score": 0.1233,
    "costPerTask": 0.0472,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2-2025-12-11-thinking-none",
    "score": 0.0083,
    "costPerTask": 0.0818,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2-2025-12-11-thinking-low",
    "score": 0.6591,
    "costPerTask": 0.1338,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2-2025-12-11-thinking-low",
    "score": 0.0833,
    "costPerTask": 0.2738,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2-2025-12-11-thinking-low",
    "score": 0.5567,
    "costPerTask": 0.1657,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2-2025-12-11-thinking-low",
    "score": 0.0972,
    "costPerTask": 0.2642,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2-2025-12-11-thinking-medium",
    "score": 0.8062,
    "costPerTask": 0.2777,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2-2025-12-11-thinking-medium",
    "score": 0.2764,
    "costPerTask": 0.7876,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2-2025-12-11-thinking-medium",
    "score": 0.7267,
    "costPerTask": 0.3448,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2-2025-12-11-thinking-medium",
    "score": 0.2667,
    "costPerTask": 0.7594,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2-2025-12-11-thinking-high",
    "score": 0.9025,
    "costPerTask": 0.402,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2-2025-12-11-thinking-high",
    "score": 0.3992,
    "costPerTask": 1.4903,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2-2025-12-11-thinking-high",
    "score": 0.7867,
    "costPerTask": 0.5196,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2-2025-12-11-thinking-high",
    "score": 0.4333,
    "costPerTask": 1.393,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2-2025-12-11-thinking-xhigh",
    "score": 0.95,
    "costPerTask": 0.65,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2-2025-12-11-thinking-xhigh",
    "score": 0.5981,
    "costPerTask": 2.054,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2-2025-12-11-thinking-xhigh",
    "score": 0.8617,
    "costPerTask": 0.96,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2-2025-12-11-thinking-xhigh",
    "score": 0.5291,
    "costPerTask": 1.8982,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2-pro-2025-12-11-medium",
    "score": 0.9025,
    "costPerTask": 3.2418,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2-pro-2025-12-11-medium",
    "score": 0.3792,
    "costPerTask": 9.5162,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2-pro-2025-12-11-medium",
    "score": 0.8117,
    "costPerTask": 3.9774,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2-pro-2025-12-11-medium",
    "score": 0.3847,
    "costPerTask": 8.9928,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2-pro-2025-12-11-high",
    "score": 0.9462,
    "costPerTask": 4.6384,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-2-pro-2025-12-11-high",
    "score": 0.5168,
    "costPerTask": 16.662,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2-pro-2025-12-11-high",
    "score": 0.8567,
    "costPerTask": 5.8694,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-2-pro-2025-12-11-high",
    "score": 0.5416,
    "costPerTask": 15.721,
    "resultsUrl": "",
    "display": true,
    "labelOffsetX": -2,
    "labelOffsetY": -10
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-2-pro-2025-12-11-xhigh",
    "score": 0.9761,
    "costPerTask": 7.7201,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-2-pro-2025-12-11-xhigh",
    "score": 0.905,
    "costPerTask": 11.6542,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-3-flash-preview-thinking-minimal",
    "score": 0.3187,
    "costPerTask": 0.0105,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-3-flash-preview-thinking-minimal",
    "score": 0.0208,
    "costPerTask": 0.0208,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-3-flash-preview-thinking-minimal",
    "score": 0.215,
    "costPerTask": 0.012,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-3-flash-preview-thinking-minimal",
    "score": 0.0333,
    "costPerTask": 0.0205,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-3-flash-preview-thinking-low",
    "score": 0.3816,
    "costPerTask": 0.0154,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-3-flash-preview-thinking-low",
    "score": 0.0125,
    "costPerTask": 0.0255,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-3-flash-preview-thinking-low",
    "score": 0.29,
    "costPerTask": 0.0163,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-3-flash-preview-thinking-low",
    "score": 0.0125,
    "costPerTask": 0.0249,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-3-flash-preview-thinking-medium",
    "score": 0.6791,
    "costPerTask": 0.0535,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-3-flash-preview-thinking-medium",
    "score": 0.1528,
    "costPerTask": 0.0825,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-3-flash-preview-thinking-medium",
    "score": 0.5767,
    "costPerTask": 0.0612,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-3-flash-preview-thinking-medium",
    "score": 0.1278,
    "costPerTask": 0.0824,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-3-flash-preview-thinking-high",
    "score": 0.8833,
    "costPerTask": 0.1353,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-3-flash-preview-thinking-high",
    "score": 0.3403,
    "costPerTask": 0.2399,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-3-flash-preview-thinking-high",
    "score": 0.8467,
    "costPerTask": 0.1743,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-3-flash-preview-thinking-high",
    "score": 0.3361,
    "costPerTask": 0.2314,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "grok-4.20-beta-0309b-reasoning",
    "score": 0.955,
    "costPerTask": 0.24,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "grok-4.20-beta-0309b-reasoning",
    "score": 0.895,
    "costPerTask": 0.35,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "grok-4.20-beta-0309b-reasoning",
    "score": 0.6361,
    "costPerTask": 0.96,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "grok-4.20-beta-0309b-reasoning",
    "score": 0.6514,
    "costPerTask": 0.92,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "johan_land_gpt_5_2_refine",
    "score": 0.945,
    "costPerTask": 11.4,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "johan_land_gpt_5_2_refine",
    "score": 0.729,
    "costPerTask": 38.99,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-6-thinking-120K-low",
    "score": 0.896,
    "costPerTask": 0.4105,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-6-thinking-120K-low",
    "score": 0.5986,
    "costPerTask": 2.3966,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-6-thinking-120K-low",
    "score": 0.86,
    "costPerTask": 0.648,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-6-thinking-120K-low",
    "score": 0.6458,
    "costPerTask": 2.2507,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-6-thinking-120K-medium",
    "score": 0.9475,
    "costPerTask": 0.6183,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-6-thinking-120K-medium",
    "score": 0.7361,
    "costPerTask": 3.0162,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-6-thinking-120K-medium",
    "score": 0.92,
    "costPerTask": 0.9647,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-6-thinking-120K-medium",
    "score": 0.6625,
    "costPerTask": 2.7209,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-6-thinking-120K-high",
    "score": 0.9625,
    "costPerTask": 0.8727,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-6-thinking-120K-high",
    "score": 0.7903,
    "costPerTask": 3.8092,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-6-thinking-120K-high",
    "score": 0.94,
    "costPerTask": 1.3948,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-6-thinking-120K-high",
    "score": 0.6917,
    "costPerTask": 3.4719,
    "resultsUrl": "",
    "display": true,
    "displayLabel": false
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-opus-4-6-thinking-120K-max",
    "score": 0.9675,
    "costPerTask": 1.2958,
    "resultsUrl": "",
    "display": true,
    "displayLabel": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-opus-4-6-thinking-120K-max",
    "score": 0.7485,
    "costPerTask": 3.7707,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-opus-4-6-thinking-120K-max",
    "score": 0.93,
    "costPerTask": 1.8791,
    "resultsUrl": "",
    "display": true,
    "labelOffsetX": -17,
    "labelOffsetY": -13
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-opus-4-6-thinking-120K-max",
    "score": 0.6875,
    "costPerTask": 3.641,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "anthropic-opus-4-8-low",
    "score": 0.88,
    "costPerTask": 0.6711,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "anthropic-opus-4-8-medium",
    "score": 0.915,
    "costPerTask": 0.9115,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "anthropic-opus-4-8-high",
    "score": 0.92,
    "costPerTask": 1.0416,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "anthropic-opus-4-8-max",
    "score": 0.925,
    "costPerTask": 2.332,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "anthropic-opus-4-8-low",
    "score": 0.6222,
    "costPerTask": 1.6818,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "anthropic-opus-4-8-medium",
    "score": 0.7167,
    "costPerTask": 2.3896,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "anthropic-opus-4-8-high",
    "score": 0.7208,
    "costPerTask": 2.7425,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini_3_deep_think_022026",
    "score": 0.96,
    "costPerTask": 7.171,
    "resultsUrl": "",
    "display": true,
    "labelOffsetX": -2,
    "labelOffsetY": -8
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini_3_deep_think_022026",
    "score": 0.8458,
    "costPerTask": 13.6213,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude_sonnet_4_6_max",
    "score": 0.9575,
    "costPerTask": 1.0867,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude_sonnet_4_6_max",
    "score": 0.6236,
    "costPerTask": 2.8905,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude_sonnet_4_6_max",
    "score": 0.86,
    "costPerTask": 1.452,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude_sonnet_4_6_max",
    "score": 0.5833,
    "costPerTask": 2.7179,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude_sonnet_4_6_high",
    "score": 0.9525,
    "costPerTask": 0.8007,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude_sonnet_4_6_high",
    "score": 0.6569,
    "costPerTask": 2.985,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude_sonnet_4_6_high",
    "score": 0.865,
    "costPerTask": 1.1908,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude_sonnet_4_6_high",
    "score": 0.6042,
    "costPerTask": 2.6966,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-3-1-pro-preview",
    "score": 0.9724,
    "costPerTask": 0.3727,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-3-1-pro-preview",
    "score": 0.8807,
    "costPerTask": 0.9789,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-3-1-pro-preview",
    "score": 0.98,
    "costPerTask": 0.5223,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-3-1-pro-preview",
    "score": 0.7708,
    "costPerTask": 0.9622,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "kimi-k2.5",
    "score": 0.7312,
    "costPerTask": 0.14,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "kimi-k2.5",
    "score": 0.6533,
    "costPerTask": 0.15,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "kimi-k2.5",
    "score": 0.1208,
    "costPerTask": 0.3,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "kimi-k2.5",
    "score": 0.1181,
    "costPerTask": 0.28,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "minimax-m2.5",
    "score": 0.5913,
    "costPerTask": 0.09,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "minimax-m2.5",
    "score": 0.6367,
    "costPerTask": 0.07,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "minimax-m2.5",
    "score": 0.0542,
    "costPerTask": 0.19,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "minimax-m2.5",
    "score": 0.0486,
    "costPerTask": 0.17,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "deepseek-v3.2",
    "score": 0.6162,
    "costPerTask": 0.07,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "deepseek-v3.2",
    "score": 0.0389,
    "costPerTask": 0.13,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "deepseek-v3.2",
    "score": 0.57,
    "costPerTask": 0.08,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "deepseek-v3.2",
    "score": 0.0403,
    "costPerTask": 0.12,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "glm-5",
    "score": 0.5863,
    "costPerTask": 0.14,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "glm-5",
    "score": 0.4467,
    "costPerTask": 0.17,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "glm-5",
    "score": 0.0542,
    "costPerTask": 0.29,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "glm-5",
    "score": 0.0486,
    "costPerTask": 0.27,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-low",
    "score": 0.8,
    "costPerTask": 0.12,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-low",
    "score": 0.6817,
    "costPerTask": 0.15,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-low",
    "score": 0.2319,
    "costPerTask": 0.29,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-low",
    "score": 0.2917,
    "costPerTask": 0.27,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-medium",
    "score": 0.92,
    "costPerTask": 0.21,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-medium",
    "score": 0.8617,
    "costPerTask": 0.25,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-medium",
    "score": 0.5819,
    "costPerTask": 0.73,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-medium",
    "score": 0.5542,
    "costPerTask": 0.68,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-high",
    "score": 0.9562,
    "costPerTask": 0.27,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-high",
    "score": 0.9267,
    "costPerTask": 0.37,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-high",
    "score": 0.7583,
    "costPerTask": 1.08,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-high",
    "score": 0.675,
    "costPerTask": 1.02,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-xhigh",
    "score": 0.9638,
    "costPerTask": 0.43,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-xhigh",
    "score": 0.9367,
    "costPerTask": 0.62,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-xhigh",
    "score": 0.8417,
    "costPerTask": 1.57,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-xhigh",
    "score": 0.7395,
    "costPerTask": 1.52,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-pro-xhigh",
    "score": 0.9825,
    "costPerTask": 5.38,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-pro-xhigh",
    "score": 0.945,
    "costPerTask": 7.21,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-pro-xhigh",
    "score": 0.9221,
    "costPerTask": 17.6,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-pro-xhigh",
    "score": 0.8333,
    "costPerTask": 16.41,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-mini-xhigh",
    "score": 0.7512,
    "costPerTask": 0.35,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-mini-xhigh",
    "score": 0.1778,
    "costPerTask": 0.81,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-mini-xhigh",
    "score": 0.6367,
    "costPerTask": 0.51,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-mini-xhigh",
    "score": 0.189,
    "costPerTask": 0.75,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-mini-high",
    "score": 0.6625,
    "costPerTask": 0.22,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-mini-high",
    "score": 0.0701,
    "costPerTask": 0.62,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-mini-high",
    "score": 0.58,
    "costPerTask": 0.27,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-mini-high",
    "score": 0.1319,
    "costPerTask": 0.56,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-mini-medium",
    "score": 0.5538,
    "costPerTask": 0.14,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-mini-medium",
    "score": 0.0542,
    "costPerTask": 0.33,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-mini-medium",
    "score": 0.4083,
    "costPerTask": 0.16,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-mini-medium",
    "score": 0.0444,
    "costPerTask": 0.33,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-mini-low",
    "score": 0.3175,
    "costPerTask": 0.03,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-mini-low",
    "score": 0.0083,
    "costPerTask": 0.06,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-mini-low",
    "score": 0.13,
    "costPerTask": 0.04,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-mini-low",
    "score": 0.0111,
    "costPerTask": 0.06,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-nano-xhigh",
    "score": 0.4788,
    "costPerTask": 0.1,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-nano-xhigh",
    "score": 0.0417,
    "costPerTask": 0.21,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-nano-xhigh",
    "score": 0.515,
    "costPerTask": 0.13,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-nano-xhigh",
    "score": 0.0569,
    "costPerTask": 0.16,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-nano-high",
    "score": 0.5162,
    "costPerTask": 0.06,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-nano-high",
    "score": 0.0514,
    "costPerTask": 0.15,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-nano-high",
    "score": 0.3817,
    "costPerTask": 0.07,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-nano-high",
    "score": 0.0361,
    "costPerTask": 0.13,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-nano-medium",
    "score": 0.4338,
    "costPerTask": 0.04,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-nano-medium",
    "score": 0.0347,
    "costPerTask": 0.06,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-nano-medium",
    "score": 0.33,
    "costPerTask": 0.04,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-nano-medium",
    "score": 0.0194,
    "costPerTask": 0.06,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-4-nano-low",
    "score": 0.2462,
    "costPerTask": 0.01,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-4-nano-low",
    "score": 0.0,
    "costPerTask": 0.02,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-4-nano-low",
    "score": 0.1833,
    "costPerTask": 0.01,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-4-nano-low",
    "score": 0.0153,
    "costPerTask": 0.01,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "anthropic-opus-4-6-max-effort",
    "score": 0.0051,
    "cost": 8866.2,
    "resultsUrl": "",
    "display": true,
    "labelOffsetX": -100,
    "labelOffsetY": -15
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "xai-grok-4-20-beta-0309-reasoning",
    "score": 0.0009,
    "cost": 3775,
    "resultsUrl": "",
    "display": true,
    "labelOffsetX": -100,
    "labelOffsetY": -15
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-4-2026-03-05-high",
    "score": 0.0021,
    "cost": 5187.41,
    "resultsUrl": "",
    "display": true,
    "labelOffsetX": -100,
    "labelOffsetY": -15
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "google-gemini-3-1-pro-preview",
    "score": 0.0042,
    "cost": 2213.53,
    "resultsUrl": "",
    "display": true,
    "labelOffsetX": -100,
    "labelOffsetY": -15
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-4-7-low",
    "score": 0.945,
    "costPerTask": 0.48,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-4-7-low",
    "score": 0.7157,
    "costPerTask": 2.62,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-4-7-low",
    "score": 0.91,
    "costPerTask": 0.76,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-4-7-low",
    "score": 0.6208,
    "costPerTask": 2.38,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-4-7-medium",
    "score": 0.9612,
    "costPerTask": 0.62,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-4-7-medium",
    "score": 0.8194,
    "costPerTask": 3.25,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-4-7-medium",
    "score": 0.91,
    "costPerTask": 1.04,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-4-7-medium",
    "score": 0.675,
    "costPerTask": 2.96,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-4-7-high",
    "score": 0.9663,
    "costPerTask": 0.84,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-4-7-high",
    "score": 0.8069,
    "costPerTask": 3.55,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-4-7-high",
    "score": 0.935,
    "costPerTask": 1.41,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-4-7-high",
    "score": 0.6833,
    "costPerTask": 3.17,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "claude-4-7-max",
    "score": 0.97,
    "costPerTask": 1.77,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "claude-4-7-max",
    "score": 0.8223,
    "costPerTask": 8.21,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "claude-4-7-max",
    "score": 0.92,
    "costPerTask": 2.58,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "claude-4-7-max",
    "score": 0.7583,
    "costPerTask": 7.43,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-5-2026-04-22-thinking-low",
    "score": 0.8225,
    "costPerTask": 0.17,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-5-2026-04-22-thinking-low",
    "score": 0.3431,
    "costPerTask": 0.37,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-5-2026-04-22-thinking-low",
    "score": 0.7617,
    "costPerTask": 0.2,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-5-2026-04-22-thinking-low",
    "score": 0.3333,
    "costPerTask": 0.35,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-5-2026-04-22-thinking-medium",
    "score": 0.965,
    "costPerTask": 0.3,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-5-2026-04-22-thinking-medium",
    "score": 0.7069,
    "costPerTask": 0.97,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-5-2026-04-22-thinking-medium",
    "score": 0.9217,
    "costPerTask": 0.39,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-5-2026-04-22-thinking-medium",
    "score": 0.7042,
    "costPerTask": 0.86,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-5-2026-04-22-thinking-high",
    "score": 0.9738,
    "costPerTask": 0.41,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-5-2026-04-22-thinking-high",
    "score": 0.8222,
    "costPerTask": 1.53,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-5-2026-04-22-thinking-high",
    "score": 0.945,
    "costPerTask": 0.56,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-5-2026-04-22-thinking-high",
    "score": 0.8333,
    "costPerTask": 1.45,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-5-2026-04-22-thinking-xhigh",
    "score": 0.975,
    "costPerTask": 0.52,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-5-2026-04-22-thinking-xhigh",
    "score": 0.9056,
    "costPerTask": 1.91,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-5-2026-04-22-thinking-xhigh",
    "score": 0.95,
    "costPerTask": 0.73,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-5-2026-04-22-thinking-xhigh",
    "score": 0.85,
    "costPerTask": 1.87,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-5-pro-2026-04-23-high",
    "score": 0.98,
    "costPerTask": 3.18,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-5-pro-2026-04-23-high",
    "score": 0.9014,
    "costPerTask": 10.44,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-5-pro-2026-04-23-high",
    "score": 0.965,
    "costPerTask": 4.53,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-5-pro-2026-04-23-high",
    "score": 0.8458,
    "costPerTask": 10.51,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gpt-5-5-pro-2026-04-23-xhigh",
    "score": 0.9799,
    "costPerTask": 3.27,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gpt-5-5-pro-2026-04-23-xhigh",
    "score": 0.9042,
    "costPerTask": 11.1,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gpt-5-5-pro-2026-04-23-xhigh",
    "score": 0.95,
    "costPerTask": 4.52,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gpt-5-5-pro-2026-04-23-xhigh",
    "score": 0.8416,
    "costPerTask": 10.76,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-5-2026-04-23-high",
    "score": 0.0043,
    "cost": 10000,
    "resultsUrl": "",
    "display": true,
    "labelOffsetX": -100,
    "labelOffsetY": -35
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "anthropic-opus-4-7-high",
    "score": 0.0018,
    "cost": 10000,
    "resultsUrl": "",
    "display": true,
    "labelOffsetX": -100,
    "labelOffsetY": -15
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "anthropic-opus-4-8-high",
    "score": 0.0152,
    "cost": 10000,
    "resultsUrl": "",
    "display": true,
    "labelOffsetX": -100,
    "labelOffsetY": -15
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-3-5-flash-minimal",
    "score": 0.58,
    "costPerTask": 0.055,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-3-5-flash-minimal",
    "score": 0.0917,
    "costPerTask": 0.1001,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-3-5-flash-minimal",
    "score": 0.4883,
    "costPerTask": 0.0645,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-3-5-flash-minimal",
    "score": 0.0889,
    "costPerTask": 0.1074,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "gemini-3-5-flash-high",
    "score": 0.96,
    "costPerTask": 0.3201,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "gemini-3-5-flash-high",
    "score": 0.7208,
    "costPerTask": 0.9001,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "gemini-3-5-flash-high",
    "score": 0.925,
    "costPerTask": 0.4278,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "gemini-3-5-flash-high",
    "score": 0.7208,
    "costPerTask": 0.8503,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-sol-low",
    "score": 0.845,
    "costPerTask": 0.14,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-low",
    "score": 0.745,
    "costPerTask": 0.17,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-sol-low",
    "score": 0.3847,
    "costPerTask": 0.33,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-low",
    "score": 0.425,
    "costPerTask": 0.32,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-sol-medium",
    "score": 0.9387,
    "costPerTask": 0.17,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-medium",
    "score": 0.925,
    "costPerTask": 0.22,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-sol-medium",
    "score": 0.7069,
    "costPerTask": 0.49,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-medium",
    "score": 0.6708,
    "costPerTask": 0.47,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-sol-high",
    "score": 0.9725,
    "costPerTask": 0.22,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-high",
    "score": 0.97,
    "costPerTask": 0.3,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-sol-high",
    "score": 0.8611,
    "costPerTask": 0.77,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-high",
    "score": 0.8542,
    "costPerTask": 0.74,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-sol-xhigh",
    "score": 0.9875,
    "costPerTask": 0.28,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-xhigh",
    "score": 0.975,
    "costPerTask": 0.4,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-sol-xhigh",
    "score": 0.9194,
    "costPerTask": 1.03,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-xhigh",
    "score": 0.9,
    "costPerTask": 1.04,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-sol-max",
    "score": 0.99,
    "costPerTask": 0.37,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-max",
    "score": 0.965,
    "costPerTask": 0.54,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-sol-max",
    "score": 0.9375,
    "costPerTask": 1.52,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-max",
    "score": 0.925,
    "costPerTask": 1.44,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-low",
    "score": 0.0033,
    "cost": 12805.6,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-medium",
    "score": 0.0107,
    "cost": 12971.17,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-high",
    "score": 0.0215,
    "cost": 15176.07,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-xhigh",
    "score": 0.0699,
    "cost": 19216.38,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-sol-max",
    "score": 0.0778,
    "cost": 25064.11,
    "resultsUrl": "/results/openai-gpt-5-6-sol",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-terra-low",
    "score": 0.7075,
    "costPerTask": 0.08,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-low",
    "score": 0.6017,
    "costPerTask": 0.09,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-terra-low",
    "score": 0.1486,
    "costPerTask": 0.17,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-low",
    "score": 0.1875,
    "costPerTask": 0.17,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-terra-medium",
    "score": 0.8475,
    "costPerTask": 0.1,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-medium",
    "score": 0.77,
    "costPerTask": 0.13,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-terra-medium",
    "score": 0.3889,
    "costPerTask": 0.28,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-medium",
    "score": 0.375,
    "costPerTask": 0.27,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-terra-high",
    "score": 0.9575,
    "costPerTask": 0.15,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-high",
    "score": 0.92,
    "costPerTask": 0.19,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-terra-high",
    "score": 0.7389,
    "costPerTask": 0.58,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-high",
    "score": 0.6708,
    "costPerTask": 0.55,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-terra-xhigh",
    "score": 0.9563,
    "costPerTask": 0.19,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-xhigh",
    "score": 0.94,
    "costPerTask": 0.26,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-terra-xhigh",
    "score": 0.8264,
    "costPerTask": 0.74,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-xhigh",
    "score": 0.7417,
    "costPerTask": 0.69,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-terra-max",
    "score": 0.9825,
    "costPerTask": 0.32,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-max",
    "score": 0.965,
    "costPerTask": 0.55,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-terra-max",
    "score": 0.9132,
    "costPerTask": 1.18,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-max",
    "score": 0.839,
    "costPerTask": 1.09,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-luna-low",
    "score": 0.4738,
    "costPerTask": 0.03,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-low",
    "score": 0.3417,
    "costPerTask": 0.04,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-luna-low",
    "score": 0.0278,
    "costPerTask": 0.07,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-low",
    "score": 0.0514,
    "costPerTask": 0.07,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-luna-medium",
    "score": 0.6438,
    "costPerTask": 0.05,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-medium",
    "score": 0.565,
    "costPerTask": 0.06,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-luna-medium",
    "score": 0.0764,
    "costPerTask": 0.11,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-medium",
    "score": 0.0736,
    "costPerTask": 0.11,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-luna-high",
    "score": 0.7925,
    "costPerTask": 0.09,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-high",
    "score": 0.765,
    "costPerTask": 0.13,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-luna-high",
    "score": 0.3014,
    "costPerTask": 0.3,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-high",
    "score": 0.2931,
    "costPerTask": 0.29,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-luna-xhigh",
    "score": 0.9,
    "costPerTask": 0.13,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-xhigh",
    "score": 0.8767,
    "costPerTask": 0.21,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-luna-xhigh",
    "score": 0.5139,
    "costPerTask": 0.53,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-xhigh",
    "score": 0.4764,
    "costPerTask": 0.54,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "openai-gpt-5-6-luna-max",
    "score": 0.9075,
    "costPerTask": 0.18,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-max",
    "score": 0.88,
    "costPerTask": 0.32,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "openai-gpt-5-6-luna-max",
    "score": 0.6017,
    "costPerTask": 0.69,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-max",
    "score": 0.5954,
    "costPerTask": 0.67,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-low",
    "score": 0.0001,
    "cost": 5068.48,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-medium",
    "score": 0.0008,
    "cost": 5895.31,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-high",
    "score": 0.0049,
    "cost": 5881.39,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-xhigh",
    "score": 0.0065,
    "cost": 6804.04,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-terra-max",
    "score": 0.008,
    "cost": 7918.31,
    "resultsUrl": "/results/openai-gpt-5-6-terra",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-low",
    "score": 0.0017,
    "cost": 2307.49,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-medium",
    "score": 0.0017,
    "cost": 2379.13,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-high",
    "score": 0.001,
    "cost": 2482.44,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-xhigh",
    "score": 0.0002,
    "cost": 2801.59,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "openai-gpt-5-6-luna-max",
    "score": 0.0018,
    "cost": 3189.42,
    "resultsUrl": "/results/openai-gpt-5-6-luna",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "glm-5.2",
    "score": 0.80375,
    "costPerTask": 0.1747,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "glm-5.2",
    "score": 0.77,
    "costPerTask": 0.1916,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "glm-5.2",
    "score": 0.2083,
    "costPerTask": 0.2294,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "glm-5.2",
    "score": 0.22775,
    "costPerTask": 0.2501,
    "resultsUrl": "",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "thinky-inkling",
    "score": 0.90625,
    "costPerTask": 0.21569280526315787,
    "resultsUrl": "/results/thinky-inkling",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "thinky-inkling",
    "score": 0.795,
    "costPerTask": 0.30047416199999993,
    "resultsUrl": "/results/thinky-inkling",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "thinky-inkling",
    "score": 0.37777777777777777,
    "costPerTask": 0.7133307604166669,
    "resultsUrl": "/results/thinky-inkling",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "thinky-inkling",
    "score": 0.36527777777777776,
    "costPerTask": 0.6420766277310923,
    "resultsUrl": "/results/thinky-inkling",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "xai-grok-4-5-high",
    "score": 0.9475,
    "costPerTask": 0.24,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "xai-grok-4-5-high",
    "score": 0.8567,
    "costPerTask": 0.33,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "xai-grok-4-5-high",
    "score": 0.5819,
    "costPerTask": 0.81,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "xai-grok-4-5-high",
    "score": 0.5264,
    "costPerTask": 0.78,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "xai-grok-4-5-medium",
    "score": 0.925,
    "costPerTask": 0.23,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "xai-grok-4-5-medium",
    "score": 0.8717,
    "costPerTask": 0.31,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "xai-grok-4-5-medium",
    "score": 0.575,
    "costPerTask": 0.75,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "xai-grok-4-5-medium",
    "score": 0.5264,
    "costPerTask": 0.75,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "xai-grok-4-5-low",
    "score": 0.86,
    "costPerTask": 0.13,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "xai-grok-4-5-low",
    "score": 0.7917,
    "costPerTask": 0.18,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "xai-grok-4-5-low",
    "score": 0.3625,
    "costPerTask": 0.4,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "xai-grok-4-5-low",
    "score": 0.3306,
    "costPerTask": 0.4,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "xai-grok-4-5-high",
    "score": 0.003,
    "cost": 6892.87,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "xai-grok-4-5-medium",
    "score": 0.0032,
    "cost": 8458.3,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "xai-grok-4-5-low",
    "score": 0.0026,
    "cost": 8718.48,
    "resultsUrl": "/results/xai-grok-4-5",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "anthropic-claude-opus-5-high",
    "score": 0.99,
    "costPerTask": 0.27,
    "resultsUrl": "/results/anthropic-claude-opus-5",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "anthropic-claude-opus-5-high",
    "score": 0.975,
    "costPerTask": 0.47,
    "resultsUrl": "/results/anthropic-claude-opus-5",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "anthropic-claude-opus-5-high",
    "score": 0.9319,
    "costPerTask": 1.32,
    "resultsUrl": "/results/anthropic-claude-opus-5",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "anthropic-claude-opus-5-high",
    "score": 0.8833,
    "costPerTask": 1.45,
    "resultsUrl": "/results/anthropic-claude-opus-5",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "anthropic-claude-opus-5-max",
    "score": 0.99,
    "costPerTask": 0.42,
    "resultsUrl": "/results/anthropic-claude-opus-5",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "anthropic-claude-opus-5-max",
    "score": 0.975,
    "costPerTask": 0.7,
    "resultsUrl": "/results/anthropic-claude-opus-5",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "anthropic-claude-opus-5-max",
    "score": 0.9706,
    "costPerTask": 2.07,
    "resultsUrl": "/results/anthropic-claude-opus-5",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "anthropic-claude-opus-5-max",
    "score": 0.9042,
    "costPerTask": 2.06,
    "resultsUrl": "/results/anthropic-claude-opus-5",
    "display": true
  },
  {
    "datasetId": "v3_Semi_Private",
    "modelId": "anthropic-claude-opus-5-high",
    "score": 0.3016,
    "cost": 20657.37,
    "resultsUrl": "/results/anthropic-claude-opus-5",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "thinky-inkling-small-xhigh",
    "score": 0.89375,
    "costPerTask": 0.08376425325000007,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "thinky-inkling-small-xhigh",
    "score": 0.84,
    "costPerTask": 0.11498211,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "thinky-inkling-small-xhigh",
    "score": 0.39305555555555555,
    "costPerTask": 0.2580068700000001,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "thinky-inkling-small-xhigh",
    "score": 0.40138888888888885,
    "costPerTask": 0.23255201249999993,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "thinky-inkling-small-high",
    "score": 0.86625,
    "costPerTask": 0.052885362000000005,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "thinky-inkling-small-high",
    "score": 0.78,
    "costPerTask": 0.06685550999999999,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "thinky-inkling-small-high",
    "score": 0.2569444444444444,
    "costPerTask": 0.14285760500000003,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "thinky-inkling-small-high",
    "score": 0.33055555555555555,
    "costPerTask": 0.132909555,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "thinky-inkling-small-medium",
    "score": 0.7775,
    "costPerTask": 0.02800050299999998,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "thinky-inkling-small-medium",
    "score": 0.67,
    "costPerTask": 0.03488058600000001,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "thinky-inkling-small-medium",
    "score": 0.1361111111111111,
    "costPerTask": 0.06787850499999998,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "thinky-inkling-small-medium",
    "score": 0.13611111111111113,
    "costPerTask": 0.06378986500000002,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "thinky-inkling-small-low",
    "score": 0.66125,
    "costPerTask": 0.021636156000000007,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "thinky-inkling-small-low",
    "score": 0.525,
    "costPerTask": 0.026465645999999992,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "thinky-inkling-small-low",
    "score": 0.06388888888888888,
    "costPerTask": 0.05634094499999999,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "thinky-inkling-small-low",
    "score": 0.04861111111111112,
    "costPerTask": 0.05589815499999999,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "thinky-inkling-small-minimal",
    "score": 0.61625,
    "costPerTask": 0.02165799,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "thinky-inkling-small-minimal",
    "score": 0.485,
    "costPerTask": 0.02370018599999999,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "thinky-inkling-small-minimal",
    "score": 0.029166666666666664,
    "costPerTask": 0.05487035499999999,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "thinky-inkling-small-minimal",
    "score": 0.04027777777777777,
    "costPerTask": 0.05303395499999999,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "thinky-inkling-small-none",
    "score": 0.11125,
    "costPerTask": 0.005578305000000001,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "thinky-inkling-small-none",
    "score": 0.055,
    "costPerTask": 0.005879189999999999,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "thinky-inkling-small-none",
    "score": 0,
    "costPerTask": 0.011877404999999999,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "thinky-inkling-small-none",
    "score": 0,
    "costPerTask": 0.010745875000000005,
    "resultsUrl": "/results/thinky-inkling-small",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "moonshot-kimi-k3-max",
    "score": 0.94875,
    "costPerTask": 0.5450828250000002,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "moonshot-kimi-k3-max",
    "score": 0.945,
    "costPerTask": 0.7704815999999995,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "moonshot-kimi-k3-max",
    "score": 0.6166666666666667,
    "costPerTask": 1.656023325000001,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "moonshot-kimi-k3-max",
    "score": 0.6041666666666666,
    "costPerTask": 1.5926940250000003,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "moonshot-kimi-k3-high",
    "score": 0.92875,
    "costPerTask": 0.36425031000000013,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "moonshot-kimi-k3-high",
    "score": 0.8666666666666666,
    "costPerTask": 0.47977101000000005,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "moonshot-kimi-k3-high",
    "score": 0.5319444444444444,
    "costPerTask": 0.9772032999999999,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "moonshot-kimi-k3-high",
    "score": 0.55,
    "costPerTask": 0.9472229249999998,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v1_Public_Eval",
    "modelId": "moonshot-kimi-k3-low",
    "score": 0.7575,
    "costPerTask": 0.16573716000000002,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v1_Semi_Private",
    "modelId": "moonshot-kimi-k3-low",
    "score": 0.6566666666666666,
    "costPerTask": 0.18061866,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v2_Public_Eval",
    "modelId": "moonshot-kimi-k3-low",
    "score": 0.1611111111111111,
    "costPerTask": 0.2610624250000001,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  },
  {
    "datasetId": "v2_Semi_Private",
    "modelId": "moonshot-kimi-k3-low",
    "score": 0.12361111111111112,
    "costPerTask": 0.2541386750000001,
    "resultsUrl": "/results/moonshot-kimi-k3",
    "display": true
  }
]