results.json

{
    "workarena_agent_curriculum": [
        {
            "Model": "GPT-3.5",
            "WorkArena-L1": 6.1,
            "WorkArena++-L2": 0.0,
            "WorkArena++-L3": 0.0,
            "MiniWoB": 43.4,
            "WebArena": 6.7
        },
        {
            "Model": "GPT-4o",
            "WorkArena-L1": 42.7,
            "WorkArena++-L2": 3.0,
            "WorkArena++-L3": 0.0,
            "MiniWoB": 71.3,
            "WebArena": 23.5
        },
        {
            "Model": "GPT-4o-V",
            "WorkArena-L1": 41.8,
            "WorkArena++-L2": 3.8,
            "WorkArena++-L3": 0.0,
            "MiniWoB": 72.5,
            "WebArena": 24.0
        },
        {
            "Model": "LLaMA-3-70b",
            "WorkArena-L1": 17.9,
            "WorkArena++-L2": 0.0,
            "WorkArena++-L3": 0.0,
            "MiniWoB": 68.2,
            "WebArena": 11.0
        },
        {
            "Model": "Mixtral-8x22b",
            "WorkArena-L1": 12.4,
            "WorkArena++-L2": 0.0,
            "WorkArena++-L3": 0.0,
            "MiniWoB": 62.4,
            "WebArena": 12.6
        }
    ],
    "workarena_l2_agent_curriculum": [
        {
            "Model": "GPT-3.5",
            "Overall": 0.0,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 0.0
        },
        {
            "Model": "GPT-4o",
            "Overall": 3.0,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 14.6
        },
        {
            "Model": "GPT-4o-V",
            "Overall": 3.8,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 3.6,
            "Sophisticated Memorization": 14.6 
        },
        {
            "Model": "LLaMA-3-70b",
            "Overall": 0.0,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 0.0
        },
        {
            "Model": "Mixtral-8x22b",
            "Overall": 0.0,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 0.0
        }
    ],
    "workarena_l2_human_curriculum": [
        {
            "Model": "Human",
            "Overall": 93.9,
            "Contextual Understanding": 100.0,
            "Data-driven Decision Making": 84.6,
            "Planning and Problem Solving": 100.0,
            "Information Retrieval": 100.0,
            "Sophisticated Memorization": 91.7
        },
        {
            "Model": "GPT-4o",
            "Overall": 2.1,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 8.3
        }
    ],
    "workarena_l3_agent_curriculum": [
        {
            "Model": "GPT-3.5",
            "Overall": 0.0,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 0.0
        },
        {
            "Model": "GPT-4o",
            "Overall": 0.0,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 0.0
        },
        {
            "Model": "GPT-4o-V",
            "Overall": 0.0,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 0.0
        },
        {
            "Model": "LLaMA-3-70b",
            "Overall": 0.0,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 0.0
        },
        {
            "Model": "Mixtral-8x22b",
            "Overall": 0.0,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 0.0
        }
    ],
    "workarena_l3_human_curriculum": [
        {
            "Model": "Human",
            "Overall": 93.9,
            "Contextual Understanding": 87.5,
            "Data-driven Decision Making": 100.0,
            "Planning and Problem Solving": 87.5,
            "Information Retrieval": 100.0,
            "Sophisticated Memorization": 91.7
        },
        {
            "Model": "GPT-4o",
            "Overall": 0.0,
            "Contextual Understanding": 0.0,
            "Data-driven Decision Making": 0.0,
            "Planning and Problem Solving": 0.0,
            "Information Retrieval": 0.0,
            "Sophisticated Memorization": 0.0
        } 
    ]
}