[
  {
    "before": "eme\":\"green\",\"strings\":{\"fs1.title\":\"FrontierScience-Olympiad accuracy across models\",\"fs1.model\":\"Model\",\"fs1.accuracy\":\"Accuracy\",\"fs1.yTitle\":\"Accuracy (%)\",\"fs1.lower\":\"Lower bound\",\"fs1.upper\":\"Upper bound\"}},\"vegaLiteSpec\":{",
    "values": [
      {
        "model": "GPT-4o",
        "accuracy_pct": 12.3,
        "ci95_low_pct": 10.860560467404067,
        "ci95_high_pct": 13.739439532595934
      },
      {
        "model": "OpenAI o1",
        "accuracy_pct": 42.5,
        "ci95_low_pct": 40.33344628499545,
        "ci95_high_pct": 44.66655371500455
      },
      {
        "model": "OpenAI o4-mini",
        "accuracy_pct": 61.650000000000006,
        "ci95_low_pct": 59.51896639585389,
        "ci95_high_pct": 63.781033604146124
      },
      {
        "model": "OpenAI o3",
        "accuracy_pct": 62.949999999999996,
        "ci95_low_pct": 60.833427998389844,
        "ci95_high_pct": 65.06657200161014
      },
      {
        "model": "Grok 4",
        "accuracy_pct": 66.2,
        "ci95_low_pct": 64.12686101575413,
        "ci95_high_pct": 68.27313898424588
      },
      {
        "model": "Claude Opus 4.5",
        "accuracy_pct": 71.35000000000001,
        "ci95_low_pct": 69.3684712613742,
        "ci95_high_pct": 73.33152873862582
      },
      {
        "model": "Gemini 3 Pro",
        "accuracy_pct": 76.1,
        "ci95_low_pct": 74.23090079664026,
        "ci95_high_pct": 77.96909920335973
      },
      {
        "model": "GPT-5.2",
        "accuracy_pct": 77.10000000000001,
        "ci95_low_pct": 75.25843944655627,
        "ci95_high_pct": 78.94156055344375
      }
    ]
  },
  {
    "before": ":{\"theme\":\"green\",\"strings\":{\"fsr.title\":\"FrontierScience-Research accuracy across models\",\"fs.model\":\"Model\",\"fs.accuracy\":\"Accuracy\",\"fs.yTitle\":\"Accuracy (%)\",\"fs.lower\":\"Lower bound\",\"fs.upper\":\"Upper bound\"}},\"vegaLiteSpec\":{",
    "values": [
      {
        "model": "GPT-4o",
        "accuracy_pct": 0.4444444444444444,
        "ci95_low_pct": 0.13714532920375716,
        "ci95_high_pct": 0.7517435596851316
      },
      {
        "model": "OpenAI o1",
        "accuracy_pct": 3.166666666666667,
        "ci95_low_pct": 2.357694769216454,
        "ci95_high_pct": 3.97563856411688
      },
      {
        "model": "OpenAI o4-mini",
        "accuracy_pct": 8.333333333333332,
        "ci95_low_pct": 7.056497932048066,
        "ci95_high_pct": 9.610168734618599
      },
      {
        "model": "Gemini 3 Pro",
        "accuracy_pct": 12.444444444444445,
        "ci95_low_pct": 10.919516126175898,
        "ci95_high_pct": 13.969372762712991
      },
      {
        "model": "OpenAI o3",
        "accuracy_pct": 14.055555555555554,
        "ci95_low_pct": 12.449898725053213,
        "ci95_high_pct": 15.661212386057894
      },
      {
        "model": "Grok 4",
        "accuracy_pct": 15.944444444444445,
        "ci95_low_pct": 14.253194769713629,
        "ci95_high_pct": 17.63569411917526
      },
      {
        "model": "Claude Opus 4.5",
        "accuracy_pct": 17.5,
        "ci95_low_pct": 15.744641537083284,
        "ci95_high_pct": 19.255358462916714
      },
      {
        "model": "GPT-5.2",
        "accuracy_pct": 25.166666666666664,
        "ci95_low_pct": 23.16182438661039,
        "ci95_high_pct": 27.171508946722938
      }
    ]
  },
  {
    "before": "s\":\"model\"},{\"calculate\":\"datum.accuracy_pct\",\"as\":\"accuracy\"},{\"calculate\":\"datum.ci95_low_pct\",\"as\":\"lower\"},{\"calculate\":\"datum.ci95_high_pct\",\"as\":\"upper\"}],\"title\":\"FrontierScience-Olympiad accuracy across reasoning efforts\",",
    "values": [
      {
        "model": "Low",
        "family": "GPT-5.2",
        "accuracy_pct": 67.45,
        "ci95_low_pct": 65.39643928796833,
        "ci95_high_pct": 69.50356071203167
      },
      {
        "model": "Medium",
        "family": "GPT-5.2",
        "accuracy_pct": 71.85000000000001,
        "ci95_low_pct": 69.8789680717959,
        "ci95_high_pct": 73.82103192820412
      },
      {
        "model": "High",
        "family": "GPT-5.2",
        "accuracy_pct": 73.75,
        "ci95_low_pct": 71.82164610353804,
        "ci95_high_pct": 75.67835389646196
      },
      {
        "model": "XHigh",
        "family": "GPT-5.2",
        "accuracy_pct": 77.10000000000001,
        "ci95_low_pct": 75.25843944655627,
        "ci95_high_pct": 78.94156055344375
      },
      {
        "model": "Low",
        "family": "OpenAI o3",
        "accuracy_pct": 53.400000000000006,
        "ci95_low_pct": 51.21372564576173,
        "ci95_high_pct": 55.586274354238284
      },
      {
        "model": "Medium",
        "family": "OpenAI o3",
        "accuracy_pct": 60.45,
        "ci95_low_pct": 58.307047868476765,
        "ci95_high_pct": 62.59295213152324
      },
      {
        "model": "High",
        "family": "OpenAI o3",
        "accuracy_pct": 62.949999999999996,
        "ci95_low_pct": 60.833427998389844,
        "ci95_high_pct": 65.06657200161014
      }
    ]
  },
  {
    "before": "um.family\"},{\"as\":\"accuracy\",\"calculate\":\"datum.accuracy_pct\"},{\"as\":\"lower\",\"calculate\":\"datum.ci95_low_pct\"},{\"as\":\"upper\",\"calculate\":\"datum.ci95_high_pct\"}],\"title\":\"FrontierScience-Research accuracy across reasoning efforts\",",
    "values": [
      {
        "model": "Low",
        "family": "GPT-5.2",
        "accuracy_pct": 18.22222222222222,
        "ci95_low_pct": 16.438865839155152,
        "ci95_high_pct": 20.00557860528929
      },
      {
        "model": "Medium",
        "family": "GPT-5.2",
        "accuracy_pct": 20.166666666666664,
        "ci95_low_pct": 18.313011225699505,
        "ci95_high_pct": 22.020322107633824
      },
      {
        "model": "High",
        "family": "GPT-5.2",
        "accuracy_pct": 21.22222222222222,
        "ci95_low_pct": 19.333286800863647,
        "ci95_high_pct": 23.111157643580796
      },
      {
        "model": "XHigh",
        "family": "GPT-5.2",
        "accuracy_pct": 25.166666666666664,
        "ci95_low_pct": 23.16182438661039,
        "ci95_high_pct": 27.171508946722938
      },
      {
        "model": "Low",
        "family": "OpenAI o3",
        "accuracy_pct": 12.88888888888889,
        "ci95_low_pct": 11.340912488704403,
        "ci95_high_pct": 14.436865289073376
      },
      {
        "model": "Medium",
        "family": "OpenAI o3",
        "accuracy_pct": 14.38888888888889,
        "ci95_low_pct": 12.767457727968946,
        "ci95_high_pct": 16.010320049808833
      },
      {
        "model": "High",
        "family": "OpenAI o3",
        "accuracy_pct": 14.055555555555554,
        "ci95_low_pct": 12.449898725053213,
        "ci95_high_pct": 15.661212386057894
      }
    ]
  }
]
