[
    {
        "model": "GPT-4",
        "Average": 65.94,
        "MMLU": 74.8,
        "WinoGrande": 66.2,
        "PiQA": 61.6,
        "CommonsenseQA": 63.0,
        "Race": 67.0,
        "MedMCQA": 51.8,
        "OpenkookQA": 60.3
    },
    {
        "model": "Claude-3 Opus",
        "Average": 62.64,
        "MMLU": 70.4,
        "WinoGrande": 63.5,
        "PiQA": 59.1,
        "CommonsenseQA": 63.7,
        "Race": 66.2,
        "MedMCQA": 49.1,
        "OpenkookQA": 54.0
    },
    {
        "model": "Mistral Large",
        "Average": 61.45,
        "MMLU": 67.8,
        "WinoGrande": 56.8,
        "PiQA": 61.2,
        "CommonsenseQA": 55.4,
        "Race": 70.1,
        "MedMCQA": 43.4,
        "OpenkookQA": 58.7
    },
    {
        "model": "GPT-3.5",
        "Average": 59.06,
        "MMLU": 65.4,
        "WinoGrande": 54.6,
        "PiQA": 54.9,
        "CommonsenseQA": 67.9,
        "Race": 60.1,
        "MedMCQA": 41.4,
        "OpenkookQA": 49.9
    },
    {
        "model": "Gemini Pro",
        "Average": 54.45,
        "MMLU": 57.7,
        "WinoGrande": 56.4,
        "PiQA": 47.7,
        "CommonsenseQA": 50.6,
        "Race": 61.0,
        "MedMCQA": 37.5,
        "OpenkookQA": 52.5
    },
    {
        "model": "Llama3-70b-instruct",
        "Average": 54.06,
        "MMLU": 64.67,
        "WinoGrande": 57.14,
        "PiQA": 43.1,
        "CommonsenseQA": 55.49,
        "Race": 58.21,
        "MedMCQA": 41.67,
        "OpenkookQA": 41.93
    }
]