mono/packages/kbot/tests/unit/reports/math.json

{
  "results": [
    {
      "test": "quadratic",
      "prompt": "Solve the quadratic equation x² + 5x + 6 = 0. Respond ONLY with the solutions as comma-separated numbers (e.g., -3,-2). No other text.",
      "result": [
        "-2,-3"
      ],
      "expected": "-3,-2",
      "model": "openai/gpt-3.5-turbo",
      "router": "openai/gpt-3.5-turbo",
      "timestamp": "2025-06-05T18:46:07.920Z",
      "passed": false,
      "duration": 1202,
      "reason": "Expected -3,-2, but got -2,-3",
      "category": "math"
    },
    {
      "test": "quadratic",
      "prompt": "Solve the quadratic equation x² + 5x + 6 = 0. Respond ONLY with the solutions as comma-separated numbers (e.g., -3,-2). No other text.",
      "result": [
        "-3,-2"
      ],
      "expected": "-3,-2",
      "model": "openai/gpt-4o-mini",
      "router": "openai/gpt-4o-mini",
      "timestamp": "2025-06-05T18:46:09.012Z",
      "passed": true,
      "duration": 1088,
      "category": "math"
    },
    {
      "test": "factorial",
      "prompt": "Calculate 5! (factorial of 5). Respond ONLY with the final numerical answer. No explanation, no other text.",
      "result": [
        "120"
      ],
      "expected": "120",
      "model": "openai/gpt-3.5-turbo",
      "router": "openai/gpt-3.5-turbo",
      "timestamp": "2025-06-05T18:46:09.517Z",
      "passed": true,
      "duration": 503,
      "category": "math"
    },
    {
      "test": "factorial",
      "prompt": "Calculate 5! (factorial of 5). Respond ONLY with the final numerical answer. No explanation, no other text.",
      "result": [
        "120"
      ],
      "expected": "120",
      "model": "openai/gpt-4o-mini",
      "router": "openai/gpt-4o-mini",
      "timestamp": "2025-06-05T18:46:10.000Z",
      "passed": true,
      "duration": 481,
      "category": "math"
    },
    {
      "test": "fibonacci",
      "prompt": "Calculate the 6th number in the Fibonacci sequence (assuming F(1)=1, F(2)=1). Respond ONLY with the final numerical answer. No other text.",
      "result": [
        "8"
      ],
      "expected": "8",
      "model": "openai/gpt-3.5-turbo",
      "router": "openai/gpt-3.5-turbo",
      "timestamp": "2025-06-05T18:46:10.505Z",
      "passed": true,
      "duration": 503,
      "category": "math"
    },
    {
      "test": "fibonacci",
      "prompt": "Calculate the 6th number in the Fibonacci sequence (assuming F(1)=1, F(2)=1). Respond ONLY with the final numerical answer. No other text.",
      "result": [
        "8"
      ],
      "expected": "8",
      "model": "openai/gpt-4o-mini",
      "router": "openai/gpt-4o-mini",
      "timestamp": "2025-06-05T18:46:11.109Z",
      "passed": true,
      "duration": 601,
      "category": "math"
    },
    {
      "test": "square_root",
      "prompt": "Calculate the square root of 16. Respond ONLY with the final numerical answer. No other text.",
      "result": [
        "4"
      ],
      "expected": "4",
      "model": "openai/gpt-3.5-turbo",
      "router": "openai/gpt-3.5-turbo",
      "timestamp": "2025-06-05T18:46:11.849Z",
      "passed": true,
      "duration": 738,
      "category": "math"
    },
    {
      "test": "square_root",
      "prompt": "Calculate the square root of 16. Respond ONLY with the final numerical answer. No other text.",
      "result": [
        "4"
      ],
      "expected": "4",
      "model": "openai/gpt-4o-mini",
      "router": "openai/gpt-4o-mini",
      "timestamp": "2025-06-05T18:46:12.390Z",
      "passed": true,
      "duration": 539,
      "category": "math"
    },
    {
      "test": "power",
      "prompt": "Calculate 2 raised to the power of 3. Respond ONLY with the final numerical answer. No other text.",
      "result": [
        "8"
      ],
      "expected": "8",
      "model": "openai/gpt-3.5-turbo",
      "router": "openai/gpt-3.5-turbo",
      "timestamp": "2025-06-05T18:46:12.985Z",
      "passed": true,
      "duration": 592,
      "category": "math"
    },
    {
      "test": "power",
      "prompt": "Calculate 2 raised to the power of 3. Respond ONLY with the final numerical answer. No other text.",
      "result": [
        "8"
      ],
      "expected": "8",
      "model": "openai/gpt-4o-mini",
      "router": "openai/gpt-4o-mini",
      "timestamp": "2025-06-05T18:46:14.091Z",
      "passed": true,
      "duration": 1103,
      "category": "math"
    }
  ],
  "highscores": [
    {
      "test": "quadratic",
      "rankings": [
        {
          "model": "openai/gpt-4o-mini",
          "duration": 1088,
          "duration_secs": 1.088
        },
        {
          "model": "openai/gpt-3.5-turbo",
          "duration": 1202,
          "duration_secs": 1.202
        }
      ]
    },
    {
      "test": "factorial",
      "rankings": [
        {
          "model": "openai/gpt-4o-mini",
          "duration": 481,
          "duration_secs": 0.481
        },
        {
          "model": "openai/gpt-3.5-turbo",
          "duration": 503,
          "duration_secs": 0.503
        }
      ]
    },
    {
      "test": "fibonacci",
      "rankings": [
        {
          "model": "openai/gpt-3.5-turbo",
          "duration": 503,
          "duration_secs": 0.503
        },
        {
          "model": "openai/gpt-4o-mini",
          "duration": 601,
          "duration_secs": 0.601
        }
      ]
    },
    {
      "test": "square_root",
      "rankings": [
        {
          "model": "openai/gpt-4o-mini",
          "duration": 539,
          "duration_secs": 0.539
        },
        {
          "model": "openai/gpt-3.5-turbo",
          "duration": 738,
          "duration_secs": 0.738
        }
      ]
    },
    {
      "test": "power",
      "rankings": [
        {
          "model": "openai/gpt-3.5-turbo",
          "duration": 592,
          "duration_secs": 0.592
        },
        {
          "model": "openai/gpt-4o-mini",
          "duration": 1103,
          "duration_secs": 1.103
        }
      ]
    }
  ],
  "lastUpdated": "2025-06-05T18:46:14.092Z"
}